See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/368395299

Driver Behavior Classification: A Systematic Literature Review

Article in IEEE Access · February 2023

DOI: 10.1109/ACCESS.2023.3243865

CITATIONS READS

18 2,835

3 authors:

Soukaina Bouhsissin Nawal Sael

Université Hassan II de Casablanca Université Hassan II de Casablanca
5 PUBLICATIONS 27 CITATIONS 96 PUBLICATIONS 627 CITATIONS

SEE PROFILE SEE PROFILE

Faouzia Benabbou
University Hassan II of Casablanca
122 PUBLICATIONS 657 CITATIONS

SEE PROFILE

All content following this page was uploaded by Soukaina Bouhsissin on 20 February 2023.

The user has requested enhancement of the downloaded file.

Received 10 January 2023, accepted 2 February 2023, date of publication 9 February 2023, date of current version 15 February 2023.
Digital Object Identifier 10.1109/ACCESS.2023.3243865

Driver Behavior Classification: A Systematic

Literature Review
SOUKAINA BOUHSISSIN , NAWAL SAEL, AND FAOUZIA BENABBOU
Laboratory of Information Technology and Modeling, Faculty of Sciences Ben M’Sick, Hassan II University of Casablanca, Casablanca 20000, Morocco
Corresponding author: Soukaina Bouhsissin ([email protected])

ABSTRACT Driver behavior is receiving increasing attention because of the staggering number of road
accidents. Many road safety reports regard human behavior as the most important factor in the likelihood
of accidents. The detection and classification of aggressive or abnormal driver behavior is an essential
requirement in the real world to avoid deadly road accidents and to protect road users. The automatic
detection of a driver’s behavior aids in the prevention of dangerous situations for the driver and all
other participants in the driving environment, as well as the implementation of corrective measures. This
paper presents a systematic literature review (SLR) of driver behavior classification. This study aimed
to highlight and analyze the different types of driver behavior, types of studies, data sources, datasets,
features, preprocessing techniques, and artificial intelligence algorithms used to classify driver behavior and
its performance. Based on the results obtained from the analysis of the selected works, we aim to identify the
key contributions and challenges of studying driver behavior classification and propose potential avenues
for further directions for practitioners and researchers.

INDEX TERMS Driver behavior, intelligent transport system, systematic literature review, machine learning,
deep learning.

I. INTRODUCTION port focuses on road safety and environmental respect, and

The intelligent transport system (ITS) concept emerged first convenience ensures service accessibility.
in 1991, when transportation experts realized that electronic ITS is now used for more than just traffic control and infor-
technologies could start to play a crucial role in optimiz- mation, but also for road and vehicle safety, infrastructure
ing surface transportation. The National ITS Program was utilization efficiency, and the reduction of accidents, injuries,
also legally established by the US Congress [1]. Intelligent and fatalities [8].
Transportation Systems are technology-based systems that Every year, road safety officials set up campaigns to raise
aim to solve various road traffic problems [2], such as traffic awareness of road accidents through interventions in schools
accidents, congestion, and conflicts, by analyzing data col- and visual and audiovisual information; additionally, public
lected from sensors or digital technology [3], [4]. ITS is a policies take measures to reduce the rate of road accidents,
general term that refers to the application of communication, for example by lowering the speed limit on the roads [9].
control, and information processing technologies to vehicle Despite these efforts, road deaths are still too high worldwide.
networks [5]. An ITS covers everything related to a transport The World Health Organization (WHO), through the Global
system, including the vehicle, infrastructure, driver, and road Status Report on Road Safety; found that an average of 3,700
users. They assist the driver in making the best decisions people died every day in traffic-related incidents in 2016,
in real time to avoid dangerous situations [5]. There are and 1.35 million traffic-related deaths occur worldwide each
three objectives of ITS: mobility, sustainable transport, and year [10]. It is therefore essential to analyze the various
convenience [6], [7]. Mobility deals with the transportation factors and driver behavior that have a direct impact on road
system’s effectiveness and capacity, while sustainable trans- accidents.
Advanced Driver Assistance Systems (ADAS) are an
The associate editor coordinating the review of this manuscript and example of in-vehicle ITS that are designed to assist drivers
approving it for publication was Wai-Keung Fung . in a variety of ways, including improving safety, reducing

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
14128 VOLUME 11, 2023
S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

the risk of accidents, and making driving more convenient.

ADAS technologies can include features such as lane depar-
ture warning systems, adaptive cruise control, automatic
emergency braking, reversing cameras, and more. These sys-
tems typically use a combination of sensors, cameras, radar,
and other technologies to gather information about the vehicle
and its surroundings, and to provide the driver with warnings,
alerts, or other forms of assistance as needed [11], [12]. How-
ever, the accuracy of these sensors can sometimes decrease
considerably, and the ADAS system cannot predict potential
danger at the right time [13].
Artificial intelligence techniques play a crucial role in
research. Machine Learning (ML) and Deep Learning (DL)
are powerful technologies that are widely used in differ-
ent contexts and applications, such as in robotics [14],
natural language processing (NLP) [15], image classifi-
cation [16], [17], and disease diagnosis like COVID-19 FIGURE 1. Driver behavior processes.
[18]. In recent years, machine learning and deep learning
techniques have accelerated the development of intelligent
transport systems (ITS) while making them more efficient, • Environmental context like traffic conditions, geometry,
especially for traffic anomaly prediction and detection [19], obstacles, and weather conditions.
accident detection and classification [16], accident severity In recent years, various commercial and research sys-
prediction [20], and other traffic problems. According to [21], tems have been proposed to analyze driver behavior and
the human factor plays a primary role with a rate of 95% develop systems to evaluate driver performance and assist
in road accidents. As a result, driver behavior analysis has drivers [26]. A common infrastructure shared by all these
become a necessity to ensure the safety of all road users.
systems is the driver surveillance system [27].
Understanding and analyzing driver behavior are essential
This paper aims to provide a systematic literature review
to identifying and addressing potential safety concerns and
(SLR) on driver behavior classification. It covers research
developing strategies to reduce the likelihood of accidents or
published between 2015 and 2022, provides a comprehen-
other road incidents. Driver behavior is a complex concept
sive overview of existing approaches to study, and analyzes
that is influenced by a range of factors, which can make it
driver behavior using artificial intelligence techniques. The
challenging to accurately describe and analyze. These factors
objective is also to offer a guide for researchers on proposed
may include psychological, social, cultural, and environmen-
approaches, performance classification techniques, datasets,
tal influences, as well as the individual characteristics of the
and features selected to classify driver behavior. The follow-
driver.
ing contains the contributions of this research:
Therefore, in parallel to ADAS and artificial intelligence,
researchers are working hard to understand, detect, identify, 1) We provided an overview of the most recent academic
and predict driving styles and behaviors. This knowledge is papers on classifying driver behavior.
crucial because driving is a common daily activity for many 2) A new taxonomy of different types of behaviors to
people [22]. By analyzing driver behavior with ML and DL classify behaviors in a systematic way, which can be
techniques, we can better understand and address the factors useful for identifying trends or patterns in behavior, and
that contribute to risky driving and work towards improving for designing interventions. In addition, an in-depth
safety on the road. DB describes the driver’s actions as they analysis to identify the various approaches that have
relate to the driving scene and the general environment [23] been used to identify driver behaviors studies.
(see Fig. 1). DB is generally evaluated in terms of environ- 3) Finally, we identify the data sources, datasets, features,
mental variables such as traffic signs, road geometry, and and preprocessing techniques such as feature extraction
pedestrians as well as vehicle variables such as distance, and selection methods, and we provide a complete
speed, acceleration, and other related variables [24]. The analysis of different ML and DL techniques involved
connections between the driver, the car, and the environment and performance obtained.
must be investigated to understand driver behavior. There- The remainder of this paper is structured as follows. Section II
fore, three contexts have to be considered [25]: presents a literature review of works related to our context.
Next, Section III presents the research methodology that we
used to select and study the papers analyzed. Section IV
• Driver context such as driver status, facial expression, presents the main results of this study in a comprehensive
and distraction. overview. Section V provides an overall discussion of the
• Car context like speed, acceleration, and orientation. main results according to the research questions. Finally,

VOLUME 11, 2023 14129

 S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

Section VI presents the conclusion of the study and its ior using smartphone sensors via an optimal path detection
limitations. algorithm and Bayesian classification. The method achieved
93.3% of correctly classified instances. For safe driving, [36]
II. RELATED WORKS suggests a system that uses two traffic datasets called Local
In general, any behavior while driving that could endanger and LARA to give drivers advice based on traffic light condi-
the car or its occupants, pedestrians, other drivers, or roadside tions. The system obtained 95.52% precision. The paper [37]
facilities may be regarded as dangerous behavior. This state proposes a smartphone-based system to provide important
of the art is organized according to the type of driver behavior information for the analysis of driver behavior at intersections
which includes five categories: Abnormal driver behavior is using a camera, an accelerometer, and gyroscope data. They
defined as unsafe behavior on the road (risky and negative); proposed Long Short-term Memory (LSTM) and Convolu-
then there’s aggressive driving, which is defined as being tional Neural Network (CNN) models and reached 0.36 for
aggressive and intolerant; line deviation, which describes the mean percentage error (MPE).
driver deviations on the road; stopping vehicles, which is a The paper [38] used the Next Generation Simulation
behavior of drivers behaving at controlled stop intersections; (NGSIM) dataset to propose an LSTM-based car tracking
and driver status, which describes driver behavior like a model that captures realistic traffic flow features and detect
distraction. In addition, we process and analyze the impact asymmetric driver behavior (which is a critical feature of
of context on driver behavior. human driver behavior). The effectiveness of road signs on
driver safety is studied in [39] using GPS and video data.
A. ABNORMAL DRIVER BEHAVIOR The Logistic Regression (LR) algorithm was trained to detect
Monitoring of abnormal driver behavior is the cornerstone visible and non-visible driver activity.
for improving driving safety. The paper [28] proposed a Some articles go further than abnormal behavior and
real-time intelligent system that can detect abnormal vehi- assessed how much the behavior presents a risk road safety.
cle behaviors using traffic cameras and the You Only Look This latter is one of the main concerns of mobility and urban
Once (YOLO) algorithm for object detection in video images. planning, so it is often important to recognize risky driver
Then the Kalman filter tracks the location of the vehicle behaviors. A Support Vector Machine (SVM) and Artificial
through successive images, and anomaly detection is carried Neural Networks (ANNs) are used to recognize safe and
out using the images in which a vehicle appears (depending dangerous driver behaviors using in-vehicle sensor data [40].
on its speed). Article [29] presents a framework based on the The classification results indicate an average accuracy of
Strategic Highway Research Program 2 (SHRP 2) and Natu- above 90% for both classifiers. The paper [41] used the SHRP
ralistic Driving Study (NDS) datasets to calculate a driver’s 2 dataset and the SVM algorithm to study the probability that
risk profile (normal or abnormal) using a Random Forest left-to-right lane changes are dangerous. The method ensured
(RF) algorithm. They were able to achieve 90% of accuracy. an accuracy of 90%. In [42], a method for detecting risky
In paper [30], the Serial-Feature Network (SF-Net) algorithm driver behaviors by analyzing vehicle speed with time in real-
was proposed for normal and abnormal driver behavior recog- world driving is proposed. The k-means and SVM algorithms
nition based on smartphone inertial sensors like GPS and achieved 95% for classes correctly classified. Papers [43]
gyroscope. The approach reached an accuracy of 97.10% and and [44] focused on the detection of driving risk levels based
a recall rate of 98.4%. A proposed system in [31] classifies on data collected by mobile sensors. The SVM model used
driver behaviors and road anomalies as normal, abnormal, in [44] performed with an accuracy higher than 70%, while
or bump. Based on smartphone sensors data were collected in [43], SVM was combined with an auto-encoder algorithm
and the k-Nearest Neighbor (KNN) and Dynamic Time Warp- and achieved 83.03% accuracy. In [45], the authors classified
ing (DTW) algorithms performed an accuracy of 78.06% and the driver’s risk level as low, medium, and high using a
96.75% respectively for classification. decision tree ‘‘CART algorithm’’ and simulator data a with
Papers [32], [33], and [34] classified driver behavior as statistical package for the social sciences (SPSS). The char-
positive and negative. The paper [32] presents a mobile appli- acteristics of driver behavior were used to assess the risk of
cation called ‘‘Project Drive’’ that bridges the gap between a vehicle-pedestrian collision based on video data [46]. This
detecting negative driver behavior and motivating users to method was archived at over 85% with high discrimination
safer driver behavior. They used the clustering k-means algo- accuracy. The paper [47] proposed an ensemble learning
rithm on GPS data. In [33], the study examines how the system for evaluating normal, low, high, and very high-risk
presence of road signs affects young drivers’ behavior in driving styles on a smartphone data using the combination
nighttime conditions using simulator and camera data. The of the following algorithms: SVM, Multi-Layer Perceptron
authors of [34] used simulator data and the MANOVA statis- (MLP), and KNN. The system succeeds in finding at least
tical technique to investigate the effects of optical circles and 94% accuracy in the driver’s style evaluation. Based on GPS
chevron patterns on driver behavior and speed when entering and ADAS data, authors in [48] suggested a system for cat-
a bend on a rural two-lane road. egorizing driving risks into low and high risks and obtained
Papers [35], [36], and [37] distinguished between safe and 80% accuracy. They focused on the unbalanced time series
unsafe driver behavior. In [35], they classified driver behav- sample problem when evaluating driving behavior, which can
14130 VOLUME 11, 2023
S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

be alleviated by MeanShift clustering. The authors of the arti- eter and gyroscope to identify aggressive driver behaviors
cle [49] proposed a real-time classification of driving behav- using statistical regression, time series analysis, and the fol-
ior based on k-means clustering, hierarchical clustering, and lowing machine learning algorithms: GMM, Partial Least
model-based clustering algorithms to identify the number Squares Regression (PLSR), wavelet transformation, and
of behavioral classes as normal, high, and low risk. Then SVR. The method achieved 77% of the F1-score using the
SVM, Decision Tree (DT), and Naive Bayes (NB) algorithms PLSR model. The paper [61] examined dangerous driving
were applied to evaluate these risk behaviors and test the events and how they are connected to traffic accidents using
performance of the clustering methods. The algorithms attend video and GPS data with correlation analysis. In [62], authors
95.3%, 99.6%, and 84.3% accuracy respectively. In another predicted the driving style of drivers based on driver activities
paper [50], they analyze the high or low skills of drivers and environmental data. Driver physiological data before and
using sensor data from a driving simulator, and an SVM during the driving start, car door opening and closing data,
algorithm was used, which performed with an accuracy of and acceleration data were collected. This approach used both
95.7 %. Additionally, [51] classifies driver behavior as skilled Bayesian Networks (BN) and Sequential Minimal Optimiza-
or non-skilled using the hidden Markov model (HMM) and tion (SMO) algorithms, with accuracy values ranging from
the model’s archived 80.37% accuracy. 72.7% to 90.9% for the aggressive driving recognition rate.
Other researchers categorize aggressive driving behav-
ior by stating the types of aggressiveness. For example,
B. AGGRESSIVE DRIVER BEHAVIOR [63] and [64] classified seven driver behaviors as aggres-
Among the most well-known driver behaviors in the lit- sive: braking, acceleration, left turn, right turn, left lane
erature is aggressive behavior. In [52], the authors deter- change, right lane change, and non-aggressive. Data were
mined whether the driving style is safe or aggressive, involv- collected from an accelerometer sensor on an Android smart-
ing signs, speed, and maneuver estimation. They used the phone, and various models were investigated, including
CNN algorithm for detection, which performed with 88.02% RNN, LSTM, and Gated Recurrent Unit (GRU) models. The
accuracy. A system based on SVM was presented in [53] experiments showed that the GRU model produced the best
to classify the various types of aggressive drivers using a accuracy results, at 95%. In [65], five driver states are clas-
few annotated simulator data points and a semi-supervised sified as driver behaviors (aggressive-stable, non-aggressive-
approach. The classification accuracy of the method was stable, non-aggressive-instable, aggressive-instable, and nor-
about 86.6% using S3VM. The paper [54] proposed a system mal) using driver behavior and EEG data. They combined k-
for normal and aggressive driver behavior classification based means, SVM, and KNN to perform an accuracy of 83.5%,
on a combination of Fully Convolutional Network (FCN) with an average accuracy of 69.5% across all tested traffic
and LSTM algorithms. Using the UAH-DriveSet dataset, the states.
system reached an F1-score of 95.88%. In [55], authors used On the other hand, some researchers were interested in
the Random Forest method to find motion-based factors that classifying the level of aggressiveness of drivers, such as
can predict aggressive driving. The model achieves a 97.10% [66], [67], [68], and [69]. They have classified drivers’ behav-
Area Under the Curve (AUC). While in the work [56], they ior with scores that express the levels of aggressiveness from
presented a significantly improved anomaly detection mech- lowest to highest using SVM and LSTM algorithms. The
anism using Recurrent Neural Networks (RNNs) based on results achieved 86.67% and 92.8% accuracy, respectively.
simulator data. This method achieves 78.6% precision and Tracking the driving style of each driver without classifica-
36.4% recall. tion was also an objective, as in the case of [70] where the
In order to identify driving maneuvers and classify aggres- authors illustrate car-following behaviors in various driving
sive, normal, and cautious driving styles, the research [57] situations using the Next Generation Simulation (NGSIM)
examines how driving habits change depending on the task dataset and genetic algorithm (GA).
performed for online car-hailing services using k-means clus-
tering. The paper [58] proposed a system for aggressive or
smooth driving style detection using the Gaussian mixture C. LINE DEVIATION
model (GMM) and data from the gyroscope. The experiment Real-time monitoring of driver events or driving style is
analyzed the driving habits of older and younger people under the cornerstone of improved driving safety. In these parts,
the same environmental tests and requirements. A super- we summarize research on line deviation detection. Papers
vised method based on Labeled Latent Dirichlet Allocation [71], [72], and [73] classified six types of driver behaviors
(LLDA) is proposed in [59] to understand driver behavior as weaving, swerving, side slipping, fast U-turn, turning with
and latent driving styles. It integrates prior knowledge via the a wide radius, and sudden braking. Acceleration and orien-
Safety Pilot Model Deployment (SPMD) dataset to classify tation data were collected, and the algorithms SVM, Neural
drivers into three categories: aggressive driving, moderate Networks (NN), and composition between SVM and NN
driving, and careful driving. The average accuracy of this were applied for classification. The models attend 95.36%,
model was 60.5%, outperforming SVM, NB, and KNN. This 96.88%, and 95.7% accuracy, respectively. In [74], authors
research [60] collected real data from the vehicle accelerom- presented a system to identify risky driving actions, such

VOLUME 11, 2023 14131

 S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

as illegal lane occupation, abrupt double lane changes, ille- tive Neuro-Fuzzy Inference System (ANFIS) were indepen-
gal U-turns, and others. Based on video surveillance data, dently designed to predict the driver’s ability to maintain
the suggested method obtains an average detection accu- the middle lane and speed limits from simulator data. The
racy of 88.62% using the hybrid algorithm Particle Swarm average error between predicted speed limit maintenance and
Optimization-Support Vector Machine (PSO SVM). In this real vehicle speed is 2.72, and the average error between
paper [75], a behavior analysis technique based on the Hidden predicted and real middle line-keeping ability is 0.27. The
Markov Model (HMM) was proposed. The aim was to assess paper [88] examines the effect of passenger presence and
the driving behavior of moving cars and identify unusual driv- driver distraction on young drivers’ behavior using simulator
ing events like approaching, braking, lane keeping, and lane data and an analysis of covariance with ANCOVA. In [89],
changing. The Conditional Monte Carlo Dense Occupancy authors proposed a system based on a Full Convolutional Net-
Tracker (CMCDOT) framework was used to determine the work (FCN) to find effective features for real-time cognitive
speed and location of nearby vehicles in real time. The results distraction detection at the wheel, and the model performed
show that the proposed method successfully detects moments with 91% for accuracy. In [90], authors presented a review to
of risk. In [76], they classified the behaviors using traffic separate and analyze the two primary categories of inattentive
data into free lane changing, non-free lane changing, suc- driving behaviors: driver distraction and driver weariness or
cessful lane changing, and unsuccessful lane changing. This drowsiness. Further, [91], [92], and [93] classified the drivers’
approach used SVM, and the prediction accuracy reached conditions as drowsy using simulator data via RF, Dynamic
nearly 90%. Bayesian Network and ANN algorithms respectively. The
results of these classifiers have an accuracy of 84.8% for RF
D. VEHICLE STOPPING and 0.22 MSE for ANN. The paper [94] exploited driving
The behavior of the driver in certain critical areas, such signals to analyze normal, aggressive, distracted, drowsy,
as stop zones is considered one of the most crucial issues and drunk driver behavior using the CNN algorithm. The
in road safety. When a yellow indication is triggered, the model achieved 99.76% accuracy. Drowsiness behavior is
dilemma zone is investigated and modeled as a binary deci- also detected in [95] and [96] using the UAH-DriveSet
sion problem to stop or go [77]. Many papers conducted dataset and LSTM algorithm. The algorithm archived 91%
binary classification of driver behavior in the dilemma zone and 99.49% of the F1-score, respectively. While in [97],
such as [78], [79], [80], [81], [82], [83]. In [78], [79], [80], they established a new model based on autoencoders for
[81], they classified the behavior using SVM, BN, Stochas- the detection of abnormal driving: drunkenness or fatigue,
tic Model Predictive Control (SMPC), and combinations of recklessness, and phone use while driving. The accuracy of
DT and Mixed Logit panel model algorithms, respectively, the model was 98.33%. Also in [98], they detected driver
that were trained with simulator data respectively. While the fatigue based on the perspective of traffic psychology. Using
SVM model predicted 92.9% accuracy, BN achieved 82.9% the SHRP 2 dataset and RF algorithms, the paper [99] catego-
precision. In a study [82], the SVM model was proposed and rizes driver behaviors as using cell phones, moving, adjusting,
trained with data collected from GPS, accelerometers, and monitoring objects, passenger interaction, talking, drinking,
sensors and achieved an accuracy of 90.02%. The video data or eating, and personal hygiene. The classifier obtains 98.5%
is used in [83] with a Binary Logistic Regression model. The concordance and a 6.5% MSE. Eating and drinking, talk-
developed model showed that the prediction accuracy of the ing, phone use while driving, and preparing are the behav-
model is 83.3%. iors classified in [100] with the interCNN algorithm. The
Understanding how drivers behave at stop-controlled inter- archived model had 81.66% accuracy. Distracted behavior
sections is of crucial importance for the control and man- is classified in papers [101], [102] as texting with the right
agement of an urban traffic system. Based on real data, the and left hands, talking on the phone using the right and
paper [84] classified driver behavior at minor street stop sign left hands, drinking, reaching, applying makeup, and talking
intersections as no-stop, rolling stop, or complete stop using to passengers using CNN based algorithms. The algorithm
binary and ordinal LR classifiers. In [85], the behavior is achieved 99% accuracy. In the paper [103], eating and tex-
classified into a full stop, slight rolling stop, ruling stop, slow ting are the two distracting behaviors classified with the RF
down without stopping, and running through stop-controlled algorithm and achieved with 85.38% accuracy. In [104], they
intersections using k-means and camera data. On the other proposed to identify the different types of secondary tasks
hand, a statistical analysis (Chi-squared test) was used to with the SHRP 2 NDS dataset in which drivers are engaged
define the types of driver behavior into complete stops, rolling in activities, such as hand-held cellphone calling, cellphone
stops, and non-compliant stops at rail level crossings (RLX) texting, and interaction. Using the combination of DT and
in the paper [86]. RF algorithms, the method achieved 99.2% accuracy at level
1 and 82.2% accuracy at level 2.
E. DRIVERS’ STATUS To further clarify the status of driving behavior, the paper
When a driver is distracted, drowsy, or has a special feeling [105] applied several different types of classifiers, such as
the driving behavior changes and affects the driving style. LR, RNN, LSTM, and Deep Neural Network (DNN), for
In [87], two non-linear regression methods, ANN and Adap- detecting driver confusion using data collected from sensors,
14132 VOLUME 11, 2023
S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

GPS, and video. In [106], a technique is proposed to robustly III. RESEARCH METHODOLOGY
classify driving styles using the Support Vector Clustering A literature review is an essential component of academic
approach into defensive, aggressive, and normal. Another research. In this review, we use the Systematic Literature
type of style is detected in [107] via a wearable glove system. Review (SLR) technique to study and analyze driver behav-
It detects driver stress events in real-time using SVM, with ior at the levels of behavioral classification types, types of
95% model accuracy. Moreover, in [108], the authors detected studies, data sources, features, preprocessing, and algorithms.
driving anger using Linear Mixed Models (LMM) and trained An SLR identifies, picks, and critically assesses research to
the model with SHRP 2 NDS dataset. The paper [109] respond to a formulated question [122]. We organized, carried
proposed a model to measure driver behavior via the driver out, and reported the review using the SLR method in [123].
behavior questionnaire (DBQ) and social media, as well as
drug and alcohol use, which were also used to measure driver A. RESEARCH QUESTIONS
behavior. The main research inquiries that need to be answered in order
to conduct the SLR for the suggested study are as follows:
F. ANALYSIS OF THE IMPACT OF CONTEXT ON DRIVER
RQ1: What are the objectives and types of driver behaviors
BEHAVIOR
classified in the selected research studies?
The objective of a driving behavior analysis is a very complex RQ2: What are the types of driver behavior studies?
issue that depends on several divergent parameters and is RQ3: What data sources, datasets, and features are used?
not restricted to a classification of driver type or driving RQ4: What data preprocessing techniques are applied to
style. Other studies focus more on analyzing the impact of improve data processing?
different factors or attributes regarding the driving environ- RQ5: What feature selection and extraction techniques are
ment on driver behavior. These factors can influence driver implemented to support the training and precision of the
behavior, cause incorrect reactions, and lead to accidents. model?
Like in [110], they studied explicit and implicit attitudes RQ6: What types of models are used to classify DB and
toward traffic climate and their relationship with drivers’ self- what is their performance?
reported behaviors. Additionally, authors in [111] examined
how driving skills affected the association between traffic
B. SEARCH STRATEGY
conditions and drivers’ behavior. To determine the extent
to which cognitive functioning contributes to the previously We use five digital databases, namely: ScienceDirect,
identified connections between driving attitudes and person- the IEEE Xplore digital library, SpringerLink, the DBLP
ality traits, the article [112] proposed an analysis of variables database, and Google Scholar. We defined a set of keywords
related to the high-risk driving behavior of young people in for the search process as ‘‘driver behavior’’, ‘‘driver behavior
the early stages of their study. Speed control is investigated classification’’, ‘‘classification of driver behavior’’, ‘‘types
of studies for driver behavior’’, ‘‘datasets for driver behavior
in [113] to analyze the different influence factors on the
classification’’, ‘‘algorithms for driver behavior’’, ‘‘data pre-
functioning and observation of right-turn drivers. In order to
processing for driver behavior’’, and we repeat the research
determine the best strategy for slowing down, [114] investi-
with the same keywords, replacing ‘‘driver behavior’’ with
gates how various perceptual treatments affect driving speed.
‘‘driver behaviour’’ as British spelling in the context of the
Another factor highlighted in [115] is the effect of longitu-
research area. Then, the search process was performed to
dinal pavement markings with varying levels and widths of
identify relevant articles to answer the search questions based
deterioration on a driver’s ability to maintain lane position.
on predefined keywords using Boolean operators in the above
Reference [116] analyzed the impact of the color contrast of
databases.
a waistcoat worn by cyclists on the visibility of drivers. Refer-
ence [117] examined whether motor vehicle drivers’ behavior
changes when there are more bicycles on the road. In addition, C. STUDY SELECTION AND QUALITY ASSESSMENT
an important factor in the analysis of driver behavior is the To choose pertinent studies, we used the inclusion and exclu-
weather condition, such as clear or foggy, which is studied in sion criteria to evaluate candidate articles that might contain
[118]. Other measures can be used to detect driver behavior information that could be used to address the research ques-
or even help avoid collisions. In [119], they seek to detect tions.
the driver’s intentions with respect to surrounding vehicles. Inclusion criteria:
In [120] the authors explored whether it is possible to iden- • The articles published from 2015 to 2022.
tify traffic congestion based on several parameters, including • The research papers are from journals publications,
delay constraints, and available speed via the GPS vehicle or conferences.
trajectory, while in [121] authors analyzed conflicts between • The works published in IEEE Xplore, DBLP, Science
vehicles and pedestrians and also driver behavior. To classify Direct, Springer and google Scholar.
the conflict into potential, mild, and severe, a model for • The articles are written in the English language.
driver yielding behavior was developed using binary logistic • The research focuses on the classification of driver
regression. behavior.

VOLUME 11, 2023 14133

 S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

FIGURE 3. Distribution of the studies over publication year.

2015-2022. Of these, 26 (28%) were published in conference

proceedings, and 67 (72%) articles were published in jour-
nals. Table 1 presents the publication venues of the chosen
studies, and Fig. 3 displays the distribution by year.

IV. DRIVER BEHAVIOR: A COMPREHENSIVE OVERVIEW

In this section, we present all the answers to the research
questions and the results concluded from the state of the art.

A. THE OBJECTIVES AND TYPES OF DRIVER BEHAVIORS

CLASSIFIED IN STUDIES (RQ1)
FIGURE 2. PRISMA flow diagram. To fully understand the types of driver behavior in literature,
we have attempted to categorize them into five categories:
abnormal driver behavior, aggressive driver behavior, stop-
• The article uses machine learning or deep learning algo- ping driver behavior, line deviation, and the driver’s status,
rithms or descriptive or preliminary statistics techniques as shown in Table 2 and Fig. 4.
in the study. • Abnormal driving behavior includes behaviors like
Exclusion criteria: abnormal, dangerous, negative, risky, high-risk, and
• The articles are not in the range of 2015 to 2022. low-risk.
• Irrelevant articles to our topic and research questions. • Aggressive driving behavior includes aggressive behav-
• We excluded duplicate articles. iors and their types and levels of aggressiveness.
• Articles that had a tenuous connection to the study’s • Vehicle stopping for driver behavior includes behaviors
questions. in intersections and dilemma zones like stop, and go.
Next, the selected articles were assessed using firstly the • Line deviation includes behaviors such as weaving,
title, abstract, conclusion, and keywords. Articles unrelated swerving, side slipping, quick u-turns, turning with a
to the contemplated study were excluded. We then applied the large radius, abrupt braking, and so on.
quality assessment criteria to the remaining articles to assess • Driver’s status such as distraction, drowsiness, stress,
their reliability, integrity, and relevance. In the review, quality use of cell phones, talking, eating, and drinking.
assessment plays an important role in the SLR protocol. The We note that the driver behaviors related to abnormal driver
selected articles perfectly answer the predefined key research behavior are studied at 30%, and the driver’s status are studied
questions based on inclusion-exclusion criteria. They are at 28%, in the selected papers. Therefore, we can deduce that
evaluated by all authors after the analysis and evaluation of the study of driver behaviors outside the car is less studied
the abstracts and conclusions of the selected articles (see and more difficult, such as the cases of line deviation 8%,
Fig. 2). and vehicle stopping 12% (see Table 2), as they are related to
the vehicle and the driving environment, such as road signs
D. DATA ANALYST and pedestrian crossings.
We identified 93 studies (Table 1) in the field of driver This work aims to understand how drivers behave on
behavior classification that were published during the period the roads, which is of crucial importance for the control,

14134 VOLUME 11, 2023

S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

TABLE 1. Publication venues.

minimization, and even avoidance of accidents; the manage- egories: Abnormal, aggressive, line deviation, stopping vehi-
ment of an urban traffic system; and many other benefits. cles, and driver status. Then, for each category, we divided
In general, each paper in Table 2 identifies and classifies these types (targets) into normal or not normal, and finally;
types of driving behavior. Some types of driver behavior we added all targets used in the articles studied.
are very common, and others are related to the study itself. We have abnormal DB present in 19.08%, which includes,
To better understand and facilitate the driver behavior types for example, abnormal, negative, unsafe, and risky. We then
analysis we regroup these types on tree levels as explained in have aggressive driver behavior and her types with 18.42%,
Fig 4. First, we categorized drive behavior types into five cat- drivers’ status with 35.53%, line deviation with 19.74%, and

VOLUME 11, 2023 14135

 S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

TABLE 2. Articles related to the type of driver behavior.

FIGURE 4. Driver behavior categorization objectives.

vehicle stopping with 7.24% as shown, in Fig. 5. Generally, behavior using instrumented vehicles; even with the use of
63 types of driver behavior are identified for the not normal monitoring equipment, instructors are frequently present in
driver behavior type (targets) and 18 types of driver behavior the vehicle to record measurements and code driving per-
for the normal targets. formance [127], [128]. The driver behavior data under the
precise control of the experiment could also be obtained in
B. DRIVER BEHAVIOR STUDIES TYPES (RQ2) the driving simulator test (driving simulator study) [129].
Research on driver behavior focuses on two aspects: objective From previous studies, we concluded that the field driving
and subjective measurement. study is most commonly used for studying driver behavior
Typically, subjective conduct measurements are derived either with cameras, sensors, etc., with 51.55% compared
based on individual viewpoints and beliefs. It represents the to the studies included in the state of the art, because it is
driver’s unique experience and is described from their point principally based on field studies, where human intervention
of view. These subjective measures can be used to evaluate is always present to determine the types of driver behavior
driver behavior through the questionnaire study. We can note and to collect data from all data sources. Followed by the
that the questionnaire survey is simple to administer and driving simulator study with 26.80%, the naturalistic driving
analyze, but it may introduce subjectivity because respon- study with 14.43%, and finally the questionnaire study with
dents occasionally fail to recognize when they engage in risky 7.22% (see Fig. 6).
behavior [124], [125]. The main problem with the first method (subjective mea-
Objective driving ratings are based on quantifiable data. sure) is that the questionnaire usually reflects the subjective
It is based on experiments via simulators or a limited sample opinions of the driver rather than the driver’s actual perfor-
of real cars. The principal sources of acknowledged objective mance on the road. The second method (objective measure)
driving data are driving simulator studies, field driving stud- usually involves manually controlling the driving environ-
ies, and naturalistic driving studies. In objective measure, the ment to induce more aggressive driver behavior. In contrast,
naturalistic driving study (NDS) uses unobtrusive measure- the second method of analysis avoids the deviation caused by
ments to record detailed information about drivers, vehicles, the drivers’ attitudes.
and their surroundings in order to study driver behavior on At the same time, in research involving naturalistic driv-
the road [126]. Field driving studies (FDS) monitor driver ing, the cars of the test subjects have equipment that, over

14136 VOLUME 11, 2023

S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

FIGURE 5. Driver behavior categorization.

TABLE 3. Data source for driver behavior classification.

FIGURE 6. Type of studies for driver behavior.

time, continuously monitors various elements of their driving

behavior in an unobtrusive manner in the absence of a test
supervisor. The strategy offers data that is challenging or
impossible to gather using existing research techniques.

C. DATA SOURCES, DATASET AND FEATURES (RQ3)

1) DATA SOURCES
To classify driver behavior, a different data sources (see
Table 3) are used to collect data. A data source is simply the data sources. From Table 4 we can classify abnormal behavior
origin of the data. It can be a file, a particular database, or even from cameras, GPS, simulators, accelerometers, and gyro-
a live data stream with sensors, cameras, etc. In our selected scope data. In addition, we can classify aggressive behavior
studies, the data source types used are smartphone sensors, from the data of simulators, smartphone sensors, gyroscopes,
GPS, accelerometers, cameras, simulators, etc. The most used GPS, etc. The vehicle’s stopping behavior is studied mainly
data source type is the simulator, with 25% of the number of from the camera and simulator data. Thus, we can classify
selected studies, followed by cameras with 17%, GPS with the driving events mostly from the sensor data (accelerometer
13%, smartphone sensors with 10%, and the other sources as and orientation sensors). The classification of the risk level is
shown in Table 3. done mainly from the sensor, simulator, and GPS data.
From the various data sources, several different types of
behavior can be extracted, depending on the researcher’s 2) DATASETS
objective and type of data source. In general, the types of To classify the driver’s behavior, some datasets are used in
driver behavior classification are associated with different the selected articles. These datasets contain a combination

VOLUME 11, 2023 14137

 S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

TABLE 4. Data source for driver behavior classification with target. TABLE 5. Datasets for driver behavior classification.

TABLE 6. Datasets for driver behavior classification with target.

of smartphone sensor data, monitoring system data, transport

agency data, and other data sources. In the following section, takes of judgment, risk-taking, aggressive driving, and traffic
we describe an example of datasets used in selected papers. violations [135].
TheStrategic Highway Research Program 2 (SHRP 2) The driver behavior dataset [63] is gathered across four
NDS database includes data from 50 million vehicle miles car excursions that last, on average, 13 minutes. The types of
and 5.4 million trips. SHRP 2 was collected by 3,100 volun- driving events and behaviors in this dataset are: aggressive
teers at six different sites in the United States: Tampa, Florida; acceleration, aggressive left and right turn, aggressive left
central Indiana; Durham, North Carolina; Erie County, New and right lane change, aggressive braking, and non-aggressive
York; central Pennsylvania; Seattle, Washington [130]. event.
TheUAH-DriveSet is a dataset used for the analysis and According to our selected paper, eight datasets are used to
classification of driving behavior. It was collected from six classify driver behavior, as shown in Table 4. The most used
distinct drivers and cars. Three separate driving behaviors, dataset is SHRP2 with 28%, followed by UAH-DriveSet with
normal, drowsy, and aggressive were included in the data 17%, as presented in Table 5.
produced [131]. In addition, Table 6 presents the targets related to each
TheNext Generation Simulation (NGSIM) Vehicle Tra- dataset. From UAH-DriveSet, we can extract normal, aggres-
jectories and Supporting Data [132] datasets are gathered sive, and drowsy driver behavior. Normal, abnormal, and
on Peachtree Street in Atlanta, Georgia, eastbound I-80 in drivers’ status (eating and drinking, personal hygiene, phone
Emeryville, California, and U.S. Highway 101 in Los Ange- use while driving, reaching, talking, talking to passengers,
les, and California. A network of synchronized digital video and texting) are extracted from SHRP2. Safe and dangerous
cameras was used to gather the data. situations are extracted from the driving dataset. More spe-
The Safety Pilot Model Deployment (SPMD) [133] cific types of aggressiveness are found in the driver behavior
includes data on driver-vehicle interactions, vehicle trajecto- dataset.
ries, basic safety messages (BSMs), and contextual data that
specifies the environment in which the model deployment 3) FEATURES
data was gathered. Over 2,700 automobiles outfitted with CV Every data source or dataset contains a set of features. Each
technology were used to collect the data between October feature or column, in the dataset, represents measurable data
2012 and April 2013. that can be used for analysis. Through this study, we catego-
The 100-Car Naturalistic Driving Study database [134] rize a set of variables or features according to the different
contains several examples of excessive driver behavior and data sources and datasets. We have grouped these features
performance, such as extreme weariness, impairment, mis- into 39 general characteristics, presented in Table 7.

14138 VOLUME 11, 2023

S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

TABLE 7. Features for driver behavior classification. source and dataset. From these figures, we can derive features
that allow drivers to describe and identify their behavior;
furthermore, these features can be exploited to create clas-
sification models of driver behavior.
Fig. 7 allows us to deduce that: The simulator helps extract
a lot of information, such as acceleration, deceleration, accel-
eration (lateral and longitudinal), throttle, direction, vehicle
position, and vehicle speed. GPS can be used to measure
the speed and acceleration of a vehicle. The accelerometer
can provide acceleration on three axes, which gives us an
accurate indication of driver behavior and road anomalies.
The gyroscope can provide the angular velocity (speed of
rotation) on the three axes (x, y, and z). Moreover, the com-
bination of vehicle speed and throttle opening can capture
the acceleration characteristics of the driver. With the camera,
we can capture images of vehicles, drivers, and surroundings
and detect the angle of rotation, vehicle speed, acceleration,
deceleration, gap, pedestrian, traffic sign, etc.
Regarding the datasets in Fig. 8, SHRP2 contains well-
known feature sets used to study driver behavior such as
acceleration (lateral and longitudinal), personal information
of drivers (gender and age), passengers, turn angle, traffic
conditions (construction zone, environmental factors, inter-
section influence, road geometry, secondary tasks, surface
condition, traffic density, traffic flow), velocity, turn signals,
vehicle speed, and other features. The 100-CAR Naturalistic
Driving Data Set contains acceleration (lateral and longi-
tudinal), vehicle heading (GPS), accelerator pedal position,
rotation rate, vehicle speed, velocity, and other features. The
UAH-DriveSet includes acceleration, distance to the vehicle
ahead, distance to the vehicle ahead in the current lane, turn
rate and turn angle, vehicle position (angle of car relative to
lane curvature, position of car relative to lane center), vehicle
speed, and other characteristics.
The ten features most commonly used to classify driver To provide additional orientation for future work and
behavior are: vehicle speed, acceleration/deceleration, rota- experimentation, we provide through this SLR for each type
tion rate, and rotation angle; pedal, throttle, and accelerator; of driver behavior the features most commonly used to
acceleration (lateral and longitudinal); time; time to an inter- describe that type. This helps researchers to choose the most
section; time to a stop line; and time to a lane crossing; traffic important features to achieve a better classification and, at the
condition; personnel information; steering; and physiological same time, link each type of driver behavior with the features
and psychological signals. Other features are also used in DB that can more efficiently describe it. Table 8 presents the most
classification, Table 7 presents these different features. commonly used features to describe the main types of driver
We find that several features describing the acceleration are behavior.
used: lateral, longitudinal, vertical, and linear acceleration. Acceleration and deceleration, vehicle speed, rotation
Lateral acceleration represents driving events such as left angle, and acceleration (lateral and longitudinal) are the fea-
turns and lane changes; longitudinal acceleration corresponds tures that should be used to classify driver behavior as aggres-
to braking and acceleration of the vehicle; vertical accelera- sive and abnormal. For vehicle stopping behavior, we need
tion represents road anomalies such as bumps and potholes; personnel information (age, driving experience, education,
and linear acceleration quantifies the force of acceleration gender, license type), vehicle speed, time to an intersection,
applied to a vehicle in all three dimensions (x, y, and z), time to the stop line, and time of day. To study and classify
excluding the force of gravity. Gravity and acceleration inputs drivers’ status, we need traffic conditions, acceleration, vehi-
from all three axes are required to calculate linear accel- cle speed, acceleration (lateral and longitudinal), the image
eration. In addition to acceleration features, several others of the driver, rotation angle, and steering. Finally, to recog-
are extracted from data sources and used many times in nize line deviation features commonly utilized are: rotation
the articles studied. The following Fig. 7 and Fig. 8 show angle, acceleration and deceleration, acceleration (lateral and
the must-used features that can be extracted from each data longitudinal), orientation, speed, gap, and velocity.

VOLUME 11, 2023 14139

 S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

FIGURE 7. Data source vs features for driver behavior classification.

FIGURE 8. Dataset vs features for driver behavior classification.

D. DATA PREPROCESSING TECHNIQUES (RQ4) data. Therefore, data preprocessing is a phase in the neces-
One of the most challenging keys to a performant machine sary data mining and analysis process that converts raw data
learning or deep learning model is the quality of the training into a format that can be comprehended and examined by

14140 VOLUME 11, 2023

S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

TABLE 8. Features related to driver behavior categorization. segmentation and imbalanced data. Fig. 9 shows that nor-
malization is the most commonly step in data preparation,
followed by cleaning and noisy data elimination and data
augmentation to improve the performance of the models.
Data resampling is a dominant step in this problem where
we have different sources of data, as well as data imbalance
processing.
• Data cleaning and remove noise
The cleaning process is required to eliminate redundant data
from the raw data, such as correlated and identical features,
duplicate rows with the same timestamp, and then substitute
missing values.
The researchers use machine learning techniques to impute
missing values such as KNN. The KNN algorithm is used
to replace missing values by approximating a point value
using the nearest points based on other features. In another
study, missing data is approximated by linear interpolation.
Because temporal data frequently suffers from noise, features
cannot be extracted directly from it since noise frequently
skews measurements of things like speed. Other research uses
statistical methods to replace missing values such as mean,
median, and standard deviation.
Position and speed errors in Controller Area Network
(CAN) buses, GPS receivers, and inertial sensors all impact
the driver behavior classification model in addition to the
measured data. Researchers use preprocessing of the raw
speed time series before training the model. Meanwhile,
others use preprocessing of the data via a smoothing filter
to remove the effect of noise. Moreover, the Kalman filter
is adopted to remove noise in some papers. Also, it’s used
to reduce measurement errors in abnormal acceleration and
deceleration values in the NGSIM dataset [38].
• Resampling stage
The sampling rate at which the traffic features are collected
is not uniform. For example, a 10 Hz camera, a 10 Hz sensor,
and a 1 Hz GPS sensor. Features can be downsampled at
a lower sampling rate or the lower-sampled features can be
upsampled at a higher sampling rate to overcome this prob-
lem. Some researchers applied an oversampling technique
based on a finite impulse response (FIR) filter well-known
in the field of signal processing research for its ability to do
algorithms. In fact, to obtain good results, the data must be oversampling (interpolation). Others solve the problem by
pre-processed to remove any impurities, solve the problems oversampling the sampled data below the highest sampling
inherent in the data, and improve its quality. frequency via linear interpolation filtering.
For a more efficient study of the various research propos- • Data normalization
als undertaken to preprocess and improve the data quality, Data normalization is an important step to achieve a model’s
we analyze the data preprocessing techniques used, the image performance, as the features have different scales. For data
and video preprocessing techniques adopted, the imbalanced normalization, researchers generally used the linear scaling
data problems, and the data labeling techniques studied sep- technique, as normalization technique for resampled time
arately. series sensor data. Another method was used for data nor-
malization by performing the FTP-72 (speed) driving test in
1) DATA PREPROCESSING advance to guarantee the results.
The data preprocessing techniques used in the state of the art • Data augmentation
are cleaning and removing noise, resampling or synchroniza- Data augmentation is frequently used as a preprocessing
tion, normalization, rolling window or data augmentation or method to help increase the size of the dataset, which in

VOLUME 11, 2023 14141

 S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

FIGURE 9. Preprocessing techniques used to classify driver behavior.

turn improves the accuracy of the model by providing a grounds were removed, and then the videos were resized to
better classification of the driver’s behavior. This technique produce lower-definition copies. This is because employing
allows extracting more information from the raw data than by high-resolution photos inherently adds storage, computing,
picking random data segments. Using a sliding window with and data transmission overheads, which would make the
a set length, the data was separated into overlapping parts. model’s design more challenging. This method also lessens
• Imbalanced data the conflicting effects of background and lighting changes on
the classification task, which can frequently occur in actual
The distribution of abnormal and normal driver behavior in driving situations. In addition, considering the optical flow
the data is generally imbalanced (impedance data), and as (OF) at the input of an image-trained model can effectively
long as the dataset has time series characteristics, most per- improve accuracy. The cubic B-spline method is used to
formant data balancing techniques are needed. Furthermore, represent the vehicle motion in the video.
additional data filtering cannot be used when processing the Image augmentation is also an important task that not only
data, as this step will not only result in the loss of temporal helps provide a diversified dataset and prevent the classifier
information but will also not fit the situation in which the from overfitting but also assists in the development of a more
algorithm has to be processed. This problem is a challenging robust classifier that could classify much more effectively.
one, and because of the data imbalance in most datasets, In most cases, manual observation models were used to
the evaluation metrics precision, recall, and F1-score are identify traffic and environmental conditions to extract fea-
introduced, nevertheless, it is not sufficiently addressed in tures that trained the model. Also, the mean RGB value is
the selected papers. To tackle this problem, some researchers used as a subtractor to center the data. The trajectories can
consider that the normalization of temporal data is sufficient be extracted from the video to solve some problems using the
and that the problem of unbalanced classes can be resolved shared source traffic vision analysis platform TvaLib.
by the algorithm itself. The MeanShift method is also used to For an imbalanced data problem, the downsampling of the
expand the samples of low-risk factors and solve the problem collected video can be solved by storing only every third
of unbalanced samples. frame, which reduces data redundancy.

2) PREPROCESSING OF VIDEOS AND IMAGES

Images and videos are important sources of driver behavior 3) DATA LABELING
data. The preprocessing of an image or video is the set of The first problem encountered in the driver behavior classifi-
operations carried out on the image that consists, on the one cation study is the labeling of the data. The data labels must
hand, of modifying the appearance of the image to extract be very accurate in order to teach the model to make correct
information more easily and, on the other hand, to remove predictions. There are many methods used to label the data:
useless information such as image noise to improve the data manually, automatically, and based on indicators in the data
quality [136]. that allow us to directly label and understand the type of driver
Analyzing video can be done by frames or by paths divided behavior.
into segments of one minute or more. In general, all of the In most cases, the data labeling is done manually, but
videos were cropped using fixed selection frames, the back- the problem becomes complex when the data increases. The

14142 VOLUME 11, 2023

S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

method requires much time and many humans to label that TABLE 9. Features extraction methods.
data.
Automatic labeling refers to any data labeling technique
that is not done by humans. This could mean labeling by
machine learning models, heuristic approaches, or a combi-
nation of both. The most commonly used technique in the
selected studies is clustering. The K-means clustering algo-
rithm is the most used one. It is based on several forms to mea-
sure similarity, like Euclidean distance and log-likelihood
distance. For example, in [53], speed was the key to labeling
driver behavior; in [42], the speed range was used; and in [65],
driving characteristics were the clustering base. Generally,
we use clustering only for continuous and categorical vari-
ables [45]. Three clustering algorithms are proposed in [49],
including k-means clustering, hierarchical clustering, and
model-based clustering, to choose the optimal number of
clusters for each and then label the data based on the speed
data. It is very important to extract static characteristics from
The other method used in the studied papers is based on the data to study driver behavior. These characteristics can be
the analysis of the datasets. They labeled aggressive behav- used to make straightforward behavioral classifications. The
ior as occurring in drivers who prefer greater longitudinal methods used for feature extraction, are mean, maximum,
acceleration and deceleration. Also, for the level of risk, they standard deviation, minimum, PCA, and others, as shown in
categorize behavior by the minimum acceleration, average Table 9. Time series data is used in papers [51] and [99], to
acceleration, and kinetic energy reduction ratio, which are extract statistical parameters of the attributes mean, standard
connected to the accident rate. The driver’s behavior at inter- deviation, maximum value, and minimum value to represent
sections is detected based on the decreasing probability of the maneuvers. The minimum and maximum values of speeds
stopping through the increasing speeds of the approaching and acceleration were used in [55]. While in [71] and [72],
vehicle and the increasing yellow duration at the signal. the statistical parameters were used such as maximum, min-
imum, the range of values, the mean, and the standard devi-
ation, and they present the main difference between the dif-
E. FEATURE SELECTION AND FEATURES EXTRACTION ferent driver behaviors. In addition, in [40], feature extraction
TECHNIQUES (RQ5) was applied based on the following basic statistical descrip-
The objective of feature selection is to find the best set tors: mean, maximum value, standard deviation, and median
of features to build performant models of the phenomena value. In [103], mean and variation are used to find effective
studied. According to previous research, the selection of trends then Predicted Error (PE) - It was calculated using
relevant features is done either automatically or manually Second-Order Taylor. Article [57] uses a threshold-based
using context knowledge. To select the most relevant fea- algorithm to extract driving maneuvers from the path. The
ture, ML learning techniques were widely used such as, point detection algorithm is used to estimate the time range
SVM [65], LR [105], and RF [48]. According to [96], accel- of the signal in search of important events [35]. The papers
eration and jerk based on timestamp and speed, which are [65], [73] use SVM to extract stability features. In [94],
provided by GPS sensors, are the most often used features in they proposed converting the signals into images using the
the classification of driver behavior. For video data in [74], recurrence plot technique. Consistent misuse of control or
authors used minimum redundancy and maximum relevance propensity to drive can be determined from the mean and
(mRMR). This approach performs better than conditional median values. The maximum value, for instance, denotes
mutual information maximization (CMIM), mutual informa- abnormal or aggressive behavior when it is higher than the
tion maximization (MIM), and ReliefF when evaluating the speed limitations that are presented. Furthermore, a fast rate
best representative trajectory histograms. A sensitivity met- of change may be indicated by the standard deviation, which
ric, mean square error (MSE), was employed to determine frequently denotes aggressive behavior. The engine is being
the bare minimum number of events or features required to driven aggressively if the standard deviation of the engine
study driver behavior. In general, few papers have proceeded speed or throttle position is high. The major properties of a
to the feature selection section. signal in the frequency domain can be captured using Hjorth
Feature extraction refers to the process of constructing parameters, which are frequently employed in feature extrac-
derived values (features) that are more informative and can tion. Hjorth activity reflects the signal’s strength, the mobility
facilitate subsequent learning steps. Two techniques were of its average frequency, and the complexity of its frequency
applied: feature engineering to create a new feature from variation [40]. The degree of dispersion and symmetry of
existing ones and dimension reduction. the data are each described using kurtosis and skewness,

VOLUME 11, 2023 14143

 S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

At the level of the dataset, Fig. 12 shows for each dataset

the classification algorithms used to predict and classify
driver behavior. LSTM is the most commonly used algorithm
for UAH-DriveSet with 11.42%. Then, SVM, RF, and DT
algorithms are often used in the SHRP2 dataset, with 5.71%
in each one. SVM and ANN are used with 2.87% in the
Driving dataset, and RNN, LSTM, and GRU are used in
Driver Behavior Dataset with 2.85% in each one. In general,
the LSTM algorithm is the most used in driver behavior
classification from the extracted dataset, with almost 17.14%,
followed by SVM with 14.29%, and RF with 11.43%, and the
other algorithms percentages are shown in Fig. 12. We can
also conclude that ML algorithms are the most commonly
FIGURE 10. Type of algorithms.
used to classify driver behavior from datasets, accounting for
51.43% of the total, while DL algorithms account for 42.86%.
In addition, from Fig. 13, SVM and LR algorithms are
respectively. Particularly, Skewness measures how much the mostly used to analyze smartphone sensor data. Also, for
data deviate from a perfectly symmetric distribution, whereas simulator data the SVM algorithm is recommend. In general,
Kurtosis assesses how heavy or light-tailed the data are in the SVM algorithm is the most used in the data source for the
comparison to a normal distribution [40]. classification of driver behavior with 19.60%, then LR and
A feature extraction procedure was required to minimize LSTM with 9.49 and 8.23%, respectively, and the percentage
the dimensionality of the feature space and obtain the most of other algorithms is shown in Fig. 13. ML algorithms
representative parameters from a large number of modi- account for 62.66% of the algorithms used for the classi-
fied features. Dimension reduction purposes in papers [29], fication of driver behavior based on different data sources,
[57], [103], [104], [106] use principal component analy- followed by the DL algorithm with 32.91% and the statistical
sis (PCA) to reduce dimensionality, improve model overfit- method with 4.43%.
ting performance, and reduce computational complexity. For
video data, the length of the vectors was fixed to reduce the 2) DRIVER BEHAVIOR CLASSIFICATION PERFORMANCE
data dimension. MEASURES
Measuring model performance is crucial to understand-
F. APPROACHES FOR DRIVER BEHAVIOR ANALYSIS (RQ6) ing and quantifying its effectiveness. Through this process,
In this section, we will identify and analyze the applications we can determine which model is the best to use for a clas-
of machine learning, deep learning, and statistical techniques sification or regression task. The Fig. 14 show that the most
and algorithms in the field of driver behavior assessment and popular performance metrics considered in the chosen studies
classification. Fig. 10 presents in percentage form the number are accuracy, F1-score, recall, and precision.
of papers that have developed ML, DL, or statistical analysis. In fact, as accuracy is the most common evaluation met-
It shows that machine learning (ML) algorithms are the most ric in classification modeling, it is adopted 48.75% of the
used; they are present in 60% of previous studies. Then deep selected studies, followed by the F1-score in 12.50% of
learning (DL) algorithms took 34.87%. Finally, statistical papers. The harmonic mean of the recall and precision values
methods are less used, with 5.13%. In this SLR, we have is represented by this metric. 11.25% used recall or sensitivity
extracted twenty machine learning (ML) algorithms used to metrics to measure the fraction of correctly classified positive
classify driver behavior. We have SVM, LR, RF, KNN, k- patterns, while 8.75% used precision metrics to measure the
means clustering, BN, DT, AdaBoost, and other algorithms fraction of correctly predicted positive patterns in a positive
presented in Fig. 11. In addition, twenty deep learning (DL) class. False-positive rate, MSE, ROC curve, and AUC are
algorithms, we have LSTM, CNN, ANN, RNN, DNN, SF- each used at 3.75%, followed by MPE at 2.5%, specificity,
Net, FCN, autoencoders, and other algorithms presented in MCC, MAE, and Concordance at 1.25% each.
Fig. 11. Six statistical techniques used, such as the T-test,
ANOVA, and ANCOVA. The algorithms SVM, LR, LSTM, 3) DRIVER BEHAVIOR CLASSIFICATION PERFORMANCE
ANN, KNN, RF, and CNN are the most commonly used, In terms of the best performing algorithm, we cannot decide
they present 49.72% of the articles studied. Fig. 11 shows the exactly which of the algorithms illustrated above (see Fig. 11)
interest that each algorithm received in previous studies. performs better than the others, as in most of the studies
selected above, their data source, features, sample size, study
1) DRIVER BEHAVIOR CLASSIFICATION ALGORITHMS environment, number of participants, metrological data, and
We have analyzed the algorithms used in the selected studies other factors are all different. To analyze the model’s per-
according to the dataset, data source, type of algorithm, and formance, we project in Table 10 the algorithm used in the
type of driver behavior study. datasets detailed in Table 5 and its measured performance.

14144 VOLUME 11, 2023

S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

FIGURE 11. Distribution of algorithms.

FIGURE 12. Distribution of driver behavior algorithms used in the dataset extracted from selected studies.

We note that in UAH-DriveSet the highest score obtained SPMD achieves an accuracy of 60.5% with the LLDA algo-
is 99.49% F1-score with the LSTM algorithm and the three rithm, and the extracted behavior types are aggressive driv-
output classifications: aggressive, drowsy, and normal. The ing, cautious driving, and moderate driving. Then 98.5%

VOLUME 11, 2023 14145

 S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

FIGURE 13. Distribution of driver behavior algorithms used in the data source extracted from selected studies.

FIGURE 15. Type of models.

FIGURE 14. Performance metrics.

concordance is obtained in the SHRP2 dataset with the RF 4) DRIVER BEHAVIOR APPROACHES
algorithm and the following outputs: eating and drinking, The algorithms identified earlier were used to assess driver
personal hygiene, phone use while driving, reaching, talk- behavior generally in three forms: (1) detecting driver behav-
ing, and talking to passengers. In addition, 90% accuracy ior; (2) predicting and classifying driver behavior; and (3)
is found in SHRP2 with both abnormal and normal behav- using statistics to study driver behavior. Fig. 15 is plotted
iors and the RF algorithm. Furthermore, the driving dataset to describe the driver behavior study techniques and models.
with the SVM algorithm, and the types of dangerous and 49.15% of studies use algorithms to predict driver behavior,
safe driver behavior obtained 90% accuracy. With the GRU followed by 32.20% that use detection models, and 18.64%
algorithm and Driver Behavior Dataset, 95% accuracy is that use statistical models.
attained, and the driver behavior is classified as aggressive In addition, in the detection models, the authors use 54%
in acceleration, braking, left and right turns, left and right ML algorithms and 44% DL algorithms. The same thing
lane changes, or non-aggressive events. Finally, the 100-CAR happens in prediction models: ML algorithms account for
Naturalistic. 64.29%, DL algorithms for 33.67%, and statistical models
Driving Data Set and the SVM algorithm found a result of for 2.04%. While in statistical models, researchers also use
90% accuracy for the types of dangerous and normal driver more ML algorithms with 57.14% and statistical algorithms
behavior. like ARIMA and others with 42.86% (see Fig. 16).

14146 VOLUME 11, 2023

S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

TABLE 10. The results of the algorithms based on the targets for each dataset.

For each type of driver behavior study, we need to know behavior (targets) are not unified and many interferences and
the most commonly chosen type of algorithm (Fig. 17). interdependencies can be detected between the terms used to
Machine learning algorithms are the most frequently used target the same goal. To attempt this goal, we extracted driver
algorithms in driving simulator studies, field driving studies, behaviors (DB) from papers studied in this SLR, analyze
naturalistic driving studies, and questionnaire studies, with the dependencies and relations between them and categorize
60.98%, 62.82%, 54.17%, and 64.29%, respectively. Then, them into abnormal DB, aggressive DB, vehicle stopping
with 37.50% in naturalistic driving studies and 34.62% in for DB, line deviation, and driver’s status. In each category,
field driving studies, deep learning algorithms are used. In the there are types of DB included like abnormal DB (normal,
questionnaire study, 35.71% presented statistical techniques. abnormal, safe, dangerous, positive, negative, high risk, etc.),
aggressive DB (aggressive behaviors and their types and
V. DISCUSSION levels of aggression), vehicle stopping (stop, run, go, rolling
The purpose of the SLR protocol used in this article is based stop, etc.), line deviation (swerving, side slipping, fast U-turn,
on the following objectives: successful lane changing, etc.), driver’s status (drowsiness,
RQ1: The classification of driver behavior is a complex stress, use of cell phone, talking, eating, etc.). An accurate
and difficult objective. It has not yet been addressed in study allows us to group and categorize the types of driver
detail in previous studies, in part because the types of driver behavior, which is an inevitable goal.

VOLUME 11, 2023 14147

 S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

FIGURE 16. Classification of model types according to algorithm types.

FIGURE 17. Classification of types of studies according to algorithm types.

RQ2: In the selected papers, four types of studies were features). Among the important features (the most frequent
extracted to classify driver behavior: questionnaire study, ones from the selected articles) are vehicle speed, acceler-
naturalistic driving study (NDS), field driving studies (FDS), ation/deceleration, rotation rate, pedal, acceleration (lateral
and driving simulator study. The questionnaire reflects the and longitudinal), time (time to the intersection, time to a stop
subjective opinions of the driver rather than the driver’s actual line, and time to a lane crossing), traffic condition, personnel
performance on the road, and the other types involve man- information, steering, and physiological and psychological
ual control of the driving environment to induce a specific signals. Due to the different sources of data, the mass of
driver behavior. In addition, field driving studies are the most data, and the different traffic characteristics, the problem of
commonly used to study driver behavior either with cameras, classification of driving behavior is more and more complex.
sensors, etc. However, due to these data, which are large and of differ-
RQ3: From this SLR, we identified the different data ent types, the research potentials in this area are promising,
sources used to extract data and classify driver behavior. especially with the development of ML, DL, and statistical
These data sources are classified in Table 3, we find: a simu- techniques.
lator, camera, GPS, smartphone sensors, accelerometer, gyro- RQ4 and RQ5:Data preprocessing techniques in the field
scope, and other data sources (Table 3). In addition, we have of driver behavior discussed in selected articles are: the clean-
synthesized datasets used in several articles, such as SHRP2 ing phase, noise removal, resampling phase, normalization,
(Strategic Highway Research Program 2), UAH-DriveSet, data augmentation, and imbalanced data. We find that these
SPMD (Safety Pilot Model Deployment), NGSIM (Next- techniques are not used extensively and sufficiently to achieve
Generation Simulation), Driver Behavior Dataset, 100-CAR the objective of driver behavior classification. For example,
Naturalistic Driving Data Set, and other datasets (Table 5). the number of features extracted in each article is too high
From these sources and datasets, several different features to allow us to specify the features that are important for
are extracted. Due to the complexity of the problem and classifying driver behavior. Furthermore, because the data
the different sources and datasets, the number of extracted in this topic is typically temporal, it requires a complex
features is abundant (39 large feature categories and 225 sub- structure and preprocessing to yield significant results. Let

14148 VOLUME 11, 2023

S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

us not forget the problem of unbalanced data, which is not the DBLP database, and Google Scholar to extract the infor-
sufficiently covered in the selected papers. In addition, data mation. This systematic review examined the literature on
labeling is a topic that most researchers do not address, and driver behavior classification from 2015 to 2022. Finally,
others do not clearly and adequately discuss. From the papers we find 93 primary empirical studies that are relevant to the
studied, we found a set of feature extraction techniques with research questions (RQs) posed in this review. The results
mathematical forms like minimum, maximum, variance, and showed that field driving studies are the most widely used
others, and using artificial intelligence algorithms like SVM, to study driver behavior classification. In addition, this SLR
RF, and PCA. At the level of feature selection, we have found states that there are many types of driver behavior classifi-
that most of the papers use the knowledge of the domain in cations. We have classified driver behavior as abnormal DB,
the choice of features. However, ML and DL techniques offer aggressive DB, vehicle stopping for DB, line deviation, and
enormous potential for feature selection, many techniques are driver status. We then identified the data sources and datasets
available for this purpose and allow for attribute selection that utilized to analyze and predict driver behavior. The Strategic
improves the performance of the models. Highway Research Program 2 Data Set (SHRP2) and the
RQ6: The SVM, LR, and LSTM algorithms are most UAH-DriveSet are the most commonly used datasets in driver
commonly used in various datasets to classify driver behavior. behavior classification. The simulator and the camera are
In general, ML algorithms are more applied in different types the most popular data sources for this problem. We explored
of studies, while DL algorithms are also less used. Therefore, preprocessing, feature selection, and feature extraction tech-
the question that arises is why machine learning algorithms niques used in the papers studied. Additionally, the SVM, LR,
are more widely used than deep learning algorithms, even and LSTM algorithms are widely used in training data to clas-
though we have data that changes over time, has a large sify driver behavior. In general, machine learning algorithms
number of features, and comes from heterogeneous data are most present in this problem of driver behavior classifi-
sources. We can say that the potential of DL techniques is cation with 60%, followed by deep learning algorithms with
not sufficiently exploited for DB study. 32.73%.
In general, we have synthesized from various literature on This study possesses some limitations in searching for
driver behavior a set of factors that relate to the external articles, as only journals and conferences indexed in Scopus
environment of the car that can influence driver behavior. between 2015 and 2022 were used. Additionally, only articles
Such as road conditions, traffic conditions, weather condi- based on a system, method, or machine and deep learn-
tions, and the presence of pedestrians, vehicles, motorcycles, ing algorithms based on driver behavior classification were
and cyclists. Beyond that, we need a global system that selected and analyzed. We concentrated mostly on articles
allows us to classify driver behavior according to the external dealing with the study of driver behavior from the outside of
environment of the car. the car, i.e., the external environment.
The use of real-time driver behavior classification in road For future research, we will first focus on studying the
vehicles has many implications for road security, for example: different data sources that allow us to extract drivers’ features
First, in-car warning systems can alert the driver to unusual in order to propose a semantic categorization of these features
driving behavior and encourage him to be cautious. Secondly, to provide more accurate description of the driver’s behavior.
we can use it to create a police warning system for abnormally Furthermore, we attempt to propose a DB classification sys-
driving vehicles by identifying, for example, vehicles that tem capable of classifying driving behavior using multiple
are likely to be dangerous or unsafe. This is achieved by data sources and leveraging the potential of deep learning
analyzing the vehicle’s distance traveled, speed, and other algorithms for time series. Finally, we need to reduce the
approaches. This system permits us to stop a person before problems of driver behavior and study how this behavior is
an accident occurs, allowing governments to impose penalties related to the environment, such as intersections, pedestrians,
on reckless drivers to maintain road safety and traffic con- and weather.
trol. Thirdly, it helps companies make better decisions about
REFERENCES
hiring drivers and predict the behavior of their employees.
[1] S. A. Shaheen and R. Finson, ‘‘Intelligent transportation systems?’’
Finally, to detect areas of abnormal behavior to assist the gov- in Reference Module in Earth Systems and Environmental
ernment in making decisions to improve road safety in those Sciences. Amsterdam, The Netherlands: Elsevier, 2016,
areas. All of these applications can help prevent accidents and Art. no. B9780124095489012000, doi: 10.1016/B978-0-12-409548-
9.01108-8.
reduce the cost of repairs. [2] N. Kumar, K. Ankit, and A. Pathani, ‘‘Intelligent transportation system:
Review paper,’’ Int. J. Sci. Eng. Res., vol. 8, no. 12, p. 5, 2017.
VI. CONCLUSION [3] J.-P. Rodrigue, C. Comtois, and B. Slack, The Geography of
Transport Systems, 3rd ed. London, U.K.: Routledge, 2013, doi:
The aim of this review is to identify existing classification
10.4324/9780203371183.
systems for driver behavior. Using SLR guidelines and proce- [4] B. Singh and A. Gupta, ‘‘Recent trends in intelligent transportation sys-
dures, we analyzed and evaluated past reviews in the field of tems: A review,’’ J. Transp. Literature, vol. 9, no. 2, pp. 30–34, Apr. 2015,
driver behavior. We followed the systematic literature review doi: 10.1590/2238-1031.jtl.v9n2a6.
[5] A. Paul, N. Chilamkurti, A. Daniel, and S. Rho, ‘‘Intelligent transporta-
approach in this study and used digital databases such as tion systems,’’ in Intelligent Vehicular Networks and Communications,
ScienceDirect, the IEEE Xplore digital library, SpringerLink, 2017, pp. 21–41, doi: 10.1016/B978-0-12-809266-8.00002-8.

VOLUME 11, 2023 14149

 S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

[6] Y. Lin, P. Wang, and M. Ma, ‘‘Intelligent transportation system (ITS): [24] M. Q. Khan and S. Lee, ‘‘A comprehensive survey of driving monitoring
Concept, challenge and opportunity,’’ in Proc. IEEE IEEE 3rd Int. Conf. and assistance systems,’’ Sensors, vol. 19, no. 11, p. 2574, Jun. 2019, doi:
Big Data Secur. Cloud (BigDataSecurity) Int. Conf. High Perform. Smart 10.3390/s19112574.
Comput., (HPSC) IEEE Int. Conf. Intell. Data Secur. (IDS), May 2017, [25] A. Soultana, F. Benabbou, and N. Sael, ‘‘Context-awareness in the smart
pp. 167–172, doi: 10.1109/BigDataSecurity.2017.50. car: Study and analysis,’’ in PervasiveHealth: Pervasive Computing Tech-
[7] J. Oskarbski, T. Marcinkowski, and M. Zawisza, ‘‘Impact of intelligent nologies for Healthcare, Oct. 2019, doi: 10.1145/3368756.3369019.
transport systems services on the level of safety and improvement of [26] H.-B. Kang, ‘‘Various approaches for driver and driving behavior mon-
traffic conditions,’’ in Smart Solutions in Today’s Transport, vol. 715, itoring: A review,’’ in Proc. IEEE Int. Conf. Comput. Vis. Workshops,
J. Mikulski, Ed. Cham, Switzerland: Springer, 2017, pp. 142–154, doi: Dec. 2013, pp. 616–623, doi: 10.1109/ICCVW.2013.85.
10.1007/978-3-319-66251-0_12. [27] L. M. Bergasa, J. Nuevo, M. A. Sotelo, R. Barea, and M. E. Lopez,
[8] M. Boltze and V. A. Tuan, ‘‘Approaches to achieve sustainability in ‘‘Real-time system for monitoring driver vigilance,’’ IEEE Trans.
traffic management,’’ Proc. Eng., vol. 142, pp. 205–212, Jan. 2016, doi: Intell. Transp. Syst., vol. 7, no. 1, pp. 63–77, Mar. 2006, doi:
10.1016/j.proeng.2016.02.033. 10.1109/TITS.2006.869598.
[9] F. B. Aghdam, H. Sadeghi-Bazargani, S. Azami-Aghdash, A. Esmaeili, [28] C. Wang, A. Musaev, P. Sheinidashtegol, and T. Atkison, ‘‘Towards detec-
H. Panahi, M. Khazaee-Pool, and M. Golestani, ‘‘Developing a national tion of abnormal vehicle behavior using traffic cameras,’’ in Big Data—
road traffic safety education program in Iran,’’ BMC Public Health, BigData 2019, vol. 11514, K. Chen, S. Seshadri, and L.-J. Zhang, Eds.
vol. 20, no. 1, p. 1064, Jul. 2020, doi: 10.1186/s12889-020-09142-1. Cham, Switzerland: Springer, 2019, pp. 125–136, doi: 10.1007/978-3-
030-23551-2_9.
[10] Global Status Report on Road Safety 2018, World Health Organization,
[29] A. E. Abdelrahman, H. S. Hassanein, and N. Abu-Ali, ‘‘Robust
Geneva, Switzerland, 2018. Accessed: Dec. 26, 2021. [Online]. Available:
data-driven framework for driver behavior profiling using supervised
https://apps.who.int/iris/handle/10665/324835
machine learning,’’ IEEE Trans. Intell. Transport. Syst., vol. 23, no. 4,
[11] P. C. Jain, ‘‘Trends in next generation intelligent transportation sys- pp. 3336–3350, Apr. 2022, doi: 10.1109/TITS.2020.3035700.
tems,’’ in Artificial Intelligence, vol. 6, M. Găiceanu, Ed. London, U.K.: [30] R. Wang, F. Xie, J. Zhao, B. Zhang, R. Sun, and J. Yang, ‘‘Smartphone
IntechOpen, 2021, doi: 10.5772/intechopen.97690. sensors-based abnormal driving behaviors detection: Serial-feature net-
[12] I. Harris, ‘‘Embedded software for automotive applications,’’ in Software work,’’ IEEE Sensors J., vol. 21, no. 14, pp. 15719–15728, Jul. 2021, doi:
Engineering for Embedded Systems, R. Oshana and M. Kraeling, Eds. 10.1109/JSEN.2020.3036862.
Oxford, U.K.: Newnes, 2013, ch. 22, pp. 767–816, doi: 10.1016/B978-0- [31] A. H. Ali, A. Atia, and M.-S.-M. Mostafa, ‘‘Recognizing driv-
12-415917-4.00022-0. ing behavior and road anomaly using smartphone sensors,’’ Int. J.
[13] S. M. Kouchak and A. Gaffar, ‘‘Estimating the driver status using long Ambient Comput. Intell., vol. 8, no. 3, pp. 22–37, Jul. 2017, doi:
short term memory,’’ in Machine Learning and Knowledge Extraction, 10.4018/IJACI.2017070102.
vol. 11713, A. Holzinger, P. Kieseberg, A. M. Tjoa, and E. Weippl, Eds. [32] K. Bahadoor and P. Hosein, ‘‘Application for the detection of dangerous
Cham, Switzerland: Springer, 2019, pp. 67–77, doi: 10.1007/978-3-030- driving and an associated gamification framework,’’ in Proc. IEEE 4th
29726-8_5. Int. Conf. Future Internet Things Cloud Workshops (FiCloudW), Vienna,
[14] Y. H. Bhosale and K. S. Patnaik, ‘‘IoT deployable lightweight deep learn- Austria, Aug. 2016, pp. 276–281, doi: 10.1109/W-FiCloud.2016.63.
ing application for COVID-19 detection with lung diseases using Rasp- [33] D. Babić, D. Babić, H. Cajner, A. Sruk, and M. Fiolić, ‘‘Effect
berryPi,’’ in Proc. Int. Conf. IoT Blockchain Technol. (ICIBT), May 2022, of road markings and traffic signs presence on young driver stress
pp. 1–6, doi: 10.1109/ICIBT52874.2022.9807725. level, eye movement and behaviour in night-time conditions: A driv-
[15] Y. Matrane, F. Benabbou, and N. Sael, ‘‘Sentiment analysis through word ing simulator study,’’ Safety, vol. 6, no. 2, p. 24, May 2020, doi:
embedding using AraBERT: Moroccan dialect use case,’’ in Proc. Int. 10.3390/safety6020024.
Conf. Digit. Age Technol. Adv. Sustain. Develop. (ICDATA), Jun. 2021, [34] H. H. Awan, A. Pirdavani, A. Houben, S. Westhof, M. Adnan, and
pp. 80–87, doi: 10.1109/ICDATA52997.2021.00024. T. Brijs, ‘‘Impact of perceptual countermeasures on driving behavior at
[16] S. Bouhsissin, N. Sael, and F. Benabbou, ‘‘Enhanced VGG19 model curves using driving simulator,’’ Traffic Injury Prevention, vol. 20, no. 1,
for accident detection and classification from video,’’ in Proc. Int. pp. 93–99, Jan. 2019, doi: 10.1080/15389588.2018.1532568.
Conf. Digit. Age Technol. Adv. Sustain. Develop. (ICDATA), Jun. 2021, [35] H. Eren, S. Makinist, E. Akin, and A. Yilmaz, ‘‘Estimating driving
pp. 39–46, doi: 10.1109/ICDATA52997.2021.00017. behavior by a smartphone,’’ in Proc. IEEE Intell. Vehicles Symp., Madrid,
[17] Y. H. Bhosale, S. Zanwar, Z. Ahmed, M. Nakrani, D. Bhuyar, and Spain, Jun. 2012, pp. 234–239, doi: 10.1109/IVS.2012.6232298.
U. Shinde, ‘‘Deep convolutional neural network based COVID-19 classi- [36] C. Bao, C. Chen, H. Kui, and X. Wang, ‘‘Safe driving at traffic lights:
fication from radiology X-ray images for IoT enabled devices,’’ in Proc. An image recognition based approach,’’ in Proc. 20th IEEE Int. Conf.
8th Int. Conf. Adv. Comput. Commun. Syst. (ICACCS), vol. 1, Mar. 2022, Mobile Data Manage. (MDM), Hong Kong, Jun. 2019, pp. 112–117, doi:
pp. 1398–1402, doi: 10.1109/ICACCS54159.2022.9785113. 10.1109/MDM.2019.00-67.
[37] Q. Wang, Y. Liu, J. Liu, Y. Gu, and S. Kamijo, ‘‘Critical areas
[18] Y. H. Bhosale and K. S. Patnaik, ‘‘Application of deep learning techniques
detection and vehicle speed estimation system towards intersection-
in diagnosis of COVID-19 (coronavirus): A systematic review,’’ Neural
related driving behavior analysis,’’ in Proc. IEEE Int. Conf. Con-
Process Lett., Sep. 2022, doi: 10.1007/s11063-022-11023-0.
sum. Electron. (ICCE), Las Vegas, NV, USA, Jan. 2018, pp. 1–6, doi:
[19] Y. Peng, C. Li, K. Wang, Z. Gao, and R. Yu, ‘‘Examining imbal- 10.1109/ICCE.2018.8326122.
anced classification algorithms in predicting real-time traffic crash risk,’’ [38] X. Huang, J. Sun, and J. Sun, ‘‘A car-following model considering asym-
Accident Anal. Prevention, vol. 144, Sep. 2020, Art. no. 105610, doi: metric driving behavior based on long short-term memory neural net-
10.1016/j.aap.2020.105610. works,’’ Transp. Res. C, Emerg. Technol., vol. 95, pp. 346–362, Oct. 2018,
[20] S. Bouhsissin, N. Sael, and F. Benabbou, ‘‘Prediction of risks in intel- doi: 10.1016/j.trc.2018.07.022.
ligent transport systems,’’ in Proc. Int. Conf. Big Data Internet Things, [39] V. Vignali, A. Bichicchi, A. Simone, C. Lantieri, G. Dondi, and M. Costa,
vol. 489, M. Lazaar, C. Duvallet, A. Touhafi, and M. A. Achhab, Eds. ‘‘Road sign vision and driver behaviour in work zones,’’ Transp.
Cham, Switzerland: Springer, 2022, pp. 303–316, doi: 10.1007/978-3- Res. F, Traffic Psychol. Behav., vol. 60, pp. 474–484, Jan. 2019, doi:
031-07969-6_23. 10.1016/j.trf.2018.11.005.
[21] L. Evans, ‘‘The dominant role of driver behavior in traffic safety,’’ [40] E. Lattanzi and V. Freschi, ‘‘Machine learning techniques to
Amer. J. Public Health, vol. 86, no. 6, pp. 784–786, Jun. 1996, doi: identify unsafe driving behavior by means of in-vehicle sensor
10.2105/AJPH.86.6.784. data,’’ Expert Syst. Appl., vol. 176, Aug. 2021, Art. no. 114818, doi:
[22] T. K. Chan, C. S. Chin, H. Chen, and X. Zhong, ‘‘A comprehensive 10.1016/j.eswa.2021.114818.
review of driver behavior analysis utilizing smartphones,’’ IEEE Trans. [41] S. Ramyar, A. Homaifar, A. Karimoddini, and E. Tunstel, ‘‘Identifica-
Intell. Transp. Syst., vol. 21, no. 10, pp. 4444–4475, Oct. 2020, doi: tion of anomalies in lane change behavior using one-class SVM,’’ in
10.1109/TITS.2019.2940481. Proc. IEEE Int. Conf. Syst., Man, Cybern. (SMC), Budapest, Hungary,
[23] C. M. Martinez, M. Heucke, B. Gao, D. Cao, and F.-Y. Wang, ‘‘Driv- Oct. 2016, pp. 4405–4410, doi: 10.1109/SMC.2016.7844924.
ing style recognition for intelligent vehicle control and advanced driver [42] D. Wang, X. Pei, L. Li, and D. Yao, ‘‘Risky driver recognition based
assistance: A survey,’’ IEEE Trans. Intell. Transp. Syst., vol. 19, no. 3, on vehicle speed time series,’’ IEEE Trans. Human-Mach. Syst., vol. 48,
pp. 666–676, Mar. 2018, doi: 10.1109/TITS.2017.2706978. no. 1, pp. 63–71, Feb. 2018, doi: 10.1109/THMS.2017.2776605.

14150 VOLUME 11, 2023

S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

[43] M. Jeon, E. Yang, E. Oh, J. Park, and C.-H. Youn, ‘‘A deterministic [62] M. Sysoev, A. Kos, J. Guna, and M. Pogačnik, ‘‘Estimation of the driving
feedback model for safe driving based on nonlinear principal analysis style based on the users’ activity and environment influence,’’ Sensors,
scheme,’’ Proc. Comput. Sci., vol. 113, pp. 454–459, Jan. 2017, doi: vol. 17, no. 10, p. 2404, Oct. 2017, doi: 10.3390/s17102404.
10.1016/j.procs.2017.08.301. [63] E. Carvalho, B. V. Ferreira, J. Ferreira, C. de Souza, H. V. Carvalho,
[44] T. Osafune, T. Takahashi, N. Kiyama, T. Sobue, H. Yamaguchi, and Y. Suhara, A. S. Pentland, and G. Pessin, ‘‘Exploiting the use of recurrent
T. Higashino, ‘‘Analysis of accident risks from driving behaviors,’’ neural networks for driver behavior profiling,’’ in Proc. Int. Joint Conf.
Int. J. Intell. Transp. Syst. Res., vol. 15, no. 3, pp. 192–202, doi: Neural Netw. (IJCNN), Anchorage, AK, USA, May 2017, pp. 3016–3021,
10.1007/s13177-016-0132-0. doi: 10.1109/IJCNN.2017.7966230.
[45] A. C. Figueira and A. P. C. Larocca, ‘‘Proposal of a driver profile [64] J. Ferreira, E. Carvalho, B. V. Ferreira, C. de Souza, Y. Suhara,
classification in relation to risk level in overtaking maneuvers,’’ Transp. A. Pentland, and G. Pessin, ‘‘Driver behavior profiling: An investiga-
Res. F, Traffic Psychol. Behav., vol. 74, pp. 375–385, Oct. 2020, doi: tion with different smartphone sensors and machine learning,’’ PLoS
10.1016/j.trf.2020.08.012. ONE, vol. 12, no. 4, Apr. 2017, Art. no. e0174959, doi: 10.1371/jour-
[46] H. Yoshitake and M. Shino, ‘‘Risk assessment based on driving behavior nal.pone.0174959.
for preventing collisions with pedestrians when making across-traffic [65] L. Yang, R. Ma, H. M. Zhang, W. Guan, and S. Jiang, ‘‘Driving behav-
turns at intersections,’’ IATSS Res., vol. 42, no. 4, pp. 240–247, Dec. 2018, ior recognition using EEG data from a simulated car-following experi-
doi: 10.1016/j.iatssr.2018.02.001. ment,’’ Accident Anal. Prevention, vol. 116, pp. 30–40, Jul. 2018, doi:
[47] M. M. Bejani and M. Ghatee, ‘‘A context aware system for driving 10.1016/j.aap.2017.11.010.
style evaluation by an ensemble learning on smartphone sensors data,’’ [66] Y. Zheng, R. T. Chase, L. Elefteriadou, V. Sisiopiku, and B. Schroeder,
Transp. Res. C, Emerg. Technol., vol. 89, pp. 303–320, Apr. 2018, doi: ‘‘Driver types and their behaviors within a high level of pedestrian activity
10.1016/j.trc.2018.02.009. environment,’’ Transp. Lett., vol. 9, no. 1, pp. 1–11, Jan. 2017, doi:
[48] H. Zhu, R. Xiao, J. Zhang, J. Liu, C. Li, and L. Yang, ‘‘A driving 10.1080/19427867.2015.1131943.
behavior risk classification framework via the unbalanced time series [67] C. Zhang, M. Patel, S. Buthpitiya, K. Lyons, B. Harrison, and
samples,’’ IEEE Trans. Instrum. Meas., vol. 71, pp. 1–12, 2022, doi: G. D. Abowd, ‘‘Driver classification based on driving behaviors,’’ in
10.1109/TIM.2022.3145359. Proc. 21st Int. Conf. Intell. User Interfaces, Sonoma, CA, USA,
[49] K. Yang, C. A. Haddad, G. Yannis, and C. Antoniou, ‘‘Classification Mar. 2016, pp. 80–84, doi: 10.1145/2856767.2856806.
and evaluation of driving behavior safety levels: A driving simulation [68] M. E. Mumcuoglu, G. Alcan, M. Unel, O. Cicek, M. Mutluergil,
study,’’ IEEE Open J. Intell. Transp. Syst., vol. 3, pp. 111–125, 2022, doi: M. Yilmaz, and K. Koprubasi, ‘‘Driving behavior classification using
10.1109/OJITS.2022.3149474. long short term memory networks,’’ in Proc. AEIT Int. Conf. Electr. Elec-
[50] N. P. Chandrasiri, K. Nawa, and A. Ishii, ‘‘Driving skill classification in tron. Technol. Automot. (AEIT AUTOMOTIVE), Turin, Italy, Jul. 2019,
curve driving scenes using machine learning,’’ J. Mod. Transp., vol. 24, pp. 1–6, doi: 10.23919/EETA.2019.8804534.
no. 3, pp. 196–206, Sep. 2016, doi: 10.1007/s40534-016-0098-2.
[69] U. E. Manawadu, T. Kawano, S. Murata, M. Kamezaki, J. Muramatsu,
[51] A. Abdelrahman, N. Abu-Ali, and H. S. Hassanein, ‘‘Driver behavior and S. Sugano, ‘‘Multiclass classification of driver perceived work-
classification in crash and near-crash events using 100-CAR naturalistic load using long short-term memory based recurrent neural network,’’
data set,’’ in Proc. IEEE Global Commun. Conf. (GLOBECOM), Singa- in Proc. IEEE Intell. Vehicles Symp. (IV), Jun. 2018, pp. 1–6, doi:
pore, Dec. 2017, pp. 1–6, doi: 10.1109/GLOCOM.2017.8253921. 10.1109/IVS.2018.8500410.
[52] M. Karaduman and H. Eren, ‘‘Deep learning based traffic direction sign
[70] J. Wang, Z. Zhang, F. Liu, and G. Lu, ‘‘Investigating heterogeneous
detection and determining driving style,’’ in Proc. Int. Conf. Comput.
car-following behaviors of different vehicle types, traffic densities and
Sci. Eng. (UBMK), Antalya, Turkey, Oct. 2017, pp. 1046–1050, doi:
road types,’’ Transp. Res. Interdiscipl. Perspect., vol. 9, Mar. 2021,
10.1109/UBMK.2017.8093453.
Art. no. 100315, doi: 10.1016/j.trip.2021.100315.
[53] W. Wang, J. Xi, A. Chong, and L. Li, ‘‘Driving style classifica-
[71] Z. Chen, J. Yu, Y. Zhu, Y. Chen, and M. Li, ‘‘D3 : Abnormal driving
tion using a semisupervised support vector machine,’’ IEEE Trans.
behaviors detection and identification using smartphone sensors,’’ in
Human-Mach. Syst., vol. 47, no. 5, pp. 650–660, Oct. 2017, doi:
Proc. 12th Annu. IEEE Int. Conf. Sens., Commun., Netw. (SECON),
10.1109/THMS.2017.2736948.
Jun. 2015, pp. 524–532, doi: 10.1109/SAHCN.2015.7338354.
[54] Y. Moukafih, H. Hafidi, and M. Ghogho, ‘‘Aggressive driving detection
using deep learning-based time series classification,’’ in Proc. IEEE Int. [72] J. Yu, Z. Chen, Y. Zhu, Y. Chen, L. Kong, and M. Li, ‘‘Fine-grained abnor-
Symp. Innov. Intell. Syst. Appl. (INISTA), Sofia, Bulgaria, Jul. 2019, mal driving behaviors detection and identification with smartphones,’’
pp. 1–5, doi: 10.1109/INISTA.2019.8778416. IEEE Trans. Mobile Comput., vol. 16, no. 8, pp. 2198–2212, Aug. 2017,
doi: 10.1109/TMC.2016.2618873.
[55] A. Jahangiri, V. J. Berardi, and S. G. Machiani, ‘‘Application of real field
connected vehicle data for aggressive driving identification on horizontal [73] B. Raj, ‘‘Monitoring and detection of abnormal driving behavior using
curves,’’ IEEE Trans. Intell. Transp. Syst., vol. 19, no. 7, pp. 2316–2324, smartphone,’’ J. Netw. Commun. Emerg. Technol., vol. 8, no. 3, pp. 1–12,
Jul. 2018, doi: 10.1109/TITS.2017.2768527. 2018. [Online]. Available: https://www.jncet.org
[56] M. Matousek, M. El-Zohairy, A. Al-Momani, F. Kargl, and C. Bosch, [74] Z. Chen, C. Wu, Z. Huang, N. Lyu, Z. Hu, M. Zhong, Y. Cheng,
‘‘Detecting anomalous driving behavior using neural networks,’’ in and B. Ran, ‘‘Dangerous driving behavior detection using video-
Proc. IEEE Intell. Vehicles Symp. (IV), Jun. 2019, pp. 2229–2235, doi: extracted vehicle trajectory histograms,’’ J. Intell. Transp. Syst.,
10.1109/IVS.2019.8814246. vol. 21, no. 5, pp. 409–421, Sep. 2017, doi: 10.1080/15472450.
[57] Y. Ma, W. Li, K. Tang, Z. Zhang, and S. Chen, ‘‘Driving style recog- 2017.1305271.
nition and comparisons among driving tasks based on driver behavior [75] C. Yang, A. Renzaglia, A. Paigwar, C. Laugier, and D. Wang, ‘‘Driv-
in the online car-hailing industry,’’ Accident Anal. Prevention, vol. 154, ing behavior assessment and anomaly detection for intelligent vehi-
May 2021, Art. no. 106096, doi: 10.1016/j.aap.2021.106096. cles,’’ in Proc. IEEE Int. Conf. Cybern. Intell. Syst. (CIS) IEEE Conf.
[58] D.-W. Koh and H.-B. Kang, ‘‘Smartphone-based modeling and detection Robot., Autom. Mechatronics (RAM), Bangkok, Thailand, Nov. 2019,
of aggressiveness reactions in senior drivers,’’ in Proc. IEEE Intell. Vehi- pp. 524–529, doi: 10.1109/CIS-RAM47153.2019.9095790.
cles Symp. (IV), Jun. 2015, pp. 12–17, doi: 10.1109/IVS.2015.7225655. [76] E.-G. Wang, J. Sun, S. Jiang, and F. Li, ‘‘Modeling the various merg-
[59] D. Chen, Z. Chen, Y. Zhang, X. Qu, M. Zhang, and C. Wu, ‘‘Driving style ing behaviors at expressway on-ramp bottlenecks using support vector
recognition under connected circumstance using a supervised hierarchical machine models,’’ Transp. Res. Proc., vol. 25, pp. 1327–1341, Jan. 2017,
Bayesian model,’’ J. Adv. Transp., vol. 2021, pp. 1–12, Jun. 2021, doi: doi: 10.1016/j.trpro.2017.05.157.
10.1155/2021/6687378. [77] D. Gazis, R. Herman, and A. Maradudin, ‘‘The problem of the amber sig-
[60] Y. Ma, Z. Zhang, S. Chen, Y. Yu, and K. Tang, ‘‘A comparative nal light in traffic flow,’’ Oper Res, vol. 8, no. 1, pp. 112–132, Feb. 1960,
study of aggressive driving behavior recognition algorithms based on doi: 10.1287/opre.8.1.112.
vehicle motion data,’’ IEEE Access, vol. 7, pp. 8028–8038, 2019, doi: [78] M. Elhenawy, A. Jahangiri, H. A. Rakha, and I. El-Shawarby, ‘‘Clas-
10.1109/ACCESS.2018.2889751. sification of driver stop/run behavior at the onset of a yellow indica-
[61] K. Min and A. Ando, ‘‘Analysis on characteristics of dangerous driving tion for different vehicles and roadway surface conditions using his-
events via recorded data of drive-recorder,’’ Transp. Res. Proc., vol. 48, torical behavior,’’ Proc. Manuf., vol. 3, pp. 858–865, Jan. 2015, doi:
pp. 1342–1363, Jan. 2020, doi: 10.1016/j.trpro.2020.08.164. 10.1016/j.promfg.2015.07.342.

VOLUME 11, 2023 14151

 S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

[79] C. Chen, Y. Chen, J. Ma, G. Zhang, and C. M. Walton, ‘‘Driver [97] J. Hu, X. Zhang, and S. Maybank, ‘‘Abnormal driving detection with
behavior formulation in intersection dilemma zones with phone normalized driving behavior data: A deep learning approach,’’ IEEE
use distraction via a Logit–Bayesian network hybrid approach,’’ Trans. Veh. Technol., vol. 69, no. 7, pp. 6943–6951, Jul. 2020, doi:
J. Intell. Transp. Syst., vol. 22, no. 4, pp. 311–324, Jul. 2018, doi: 10.1109/TVT.2020.2993247.
10.1080/15472450.2017.1350921. [98] Y. Fan, ‘‘Detection of driver fatigue driving behaviour based on
[80] W. Li, L. Tan, and C. Lin, ‘‘Modeling driver behavior in the dilemma zone deep learning algorithm from the perspective of traffic psychol-
based on stochastic model predictive control,’’ PLoS ONE, vol. 16, no. 2, ogy,’’ J. Psychol. Afr., vol. 32, no. 3, pp. 303–309, May 2022, doi:
Feb. 2021, Art. no. e0247453, doi: 10.1371/journal.pone.0247453. 10.1080/14330237.2022.2075622.
[81] Y. Ali, M. M. Haque, Z. Zheng, and M. C. J. Bliemer, ‘‘Stop or go [99] B. Lv, R. Yue, and Y. Zhang, ‘‘The influence of different factors on
decisions at the onset of yellow light in a connected environment: right-turn distracted driving behavior at intersections using naturalistic
A hybrid approach of decision tree and panel mixed Logit model,’’ driving study data,’’ IEEE Access, vol. 7, pp. 137241–137250, 2019, doi:
Anal. Methods Accident Res., vol. 31, Sep. 2021, Art. no. 100165, doi: 10.1109/ACCESS.2019.2942841.
10.1016/j.amar.2021.100165. [100] C. Zhang, R. Li, W. Kim, D. Yoon, and P. Patras, ‘‘Driver behavior
[82] M. Elhenawy, A. Jahangiri, H. A. Rakha, and I. El-Shawarby, ‘‘Modeling recognition via interwoven deep convolutional neural nets with multi-
driver stop/run behavior at the onset of a yellow indication considering stream inputs,’’ IEEE Access, vol. 8, pp. 191138–191151, 2020, doi:
driver run tendency and roadway surface conditions,’’ Accident Anal. Pre- 10.1109/ACCESS.2020.3032344.
vention, vol. 83, pp. 90–100, Oct. 2015, doi: 10.1016/j.aap.2015.06.016. [101] S. Masood, A. Rai, A. Aggarwal, M. N. Doja, and M. Ahmad,
[83] B. K. Pathivada and V. Perumal, ‘‘Modeling driver behavior in dilemma ‘‘Detecting distraction of drivers using convolutional neural net-
zone under mixed traffic conditions,’’ Transp. Res. Proc., vol. 27, work,’’ Pattern Recognit. Lett., vol. 139, pp. 79–85, Nov. 2020, doi:
pp. 961–968, Jan. 2017, doi: 10.1016/j.trpro.2017.12.120. 10.1016/j.patrec.2017.12.023.
[84] K. Shaaban, J. S. Wood, and V. V. Gayah, ‘‘Investigating driver behavior [102] W. Huang, X. Liu, M. Luo, P. Zhang, W. Wang, and J. Wang,
at minor-street stop-controlled intersections in Qatar,’’ Transp. Res. Rec., ‘‘Video-based abnormal driving behavior detection via deep learn-
J. Transp. Res. Board, vol. 2663, no. 1, pp. 109–116, Jan. 2017, doi: ing fusions,’’ IEEE Access, vol. 7, pp. 64571–64582, 2019, doi:
10.3141/2663-14. 10.1109/ACCESS.2019.2917213.
[85] X. Wen, L. Fu, T. Fu, J. Keung, and M. Zhong, ‘‘Driver behavior [103] M. Atiquzzaman, Y. Qi, and R. Fries, ‘‘Real-time detection of drivers’
classification at stop-controlled intersections using video-based trajec- texting and eating behavior based on vehicle dynamics,’’ Transp.
tory data,’’ Sustainability, vol. 13, no. 3, pp. 1–18, Feb. 2021, doi: Res. F, Traffic Psychol. Behav., vol. 58, pp. 594–604, Oct. 2018, doi:
10.3390/su13031404. 10.1016/j.trf.2018.06.027.
[86] V. Beanland, P. M. Salmon, A. J. Filtness, M. G. Lenné, and [104] O. A. Osman, M. Hajij, S. Karbalaieali, and S. Ishak, ‘‘A hierar-
N. A. Stanton, ‘‘To stop or not to stop: Contrasting compliant and chical machine learning classification approach for secondary task
non-compliant driver behaviour at rural rail level crossings,’’ Acci- identification from observed driving behavior data,’’ Accident Anal.
dent Anal. Prevention, vol. 108, pp. 209–219, Nov. 2017, doi: Prevention, vol. 123, pp. 274–281, Feb. 2019, doi: 10.1016/j.aap.
10.1016/j.aap.2017.09.004. 2018.12.005.
[87] A. Aksjonov, P. Nedoma, V. Vodovozov, E. Petlenkov, and M. Herrmann, [105] C. Hori, S. Watanabe, T. Hori, B. A. Harsham, J. Hershey, Y. Koji, Y. Fujii,
‘‘A novel driver performance model based on machine learning,’’ and Y. Furumoto, ‘‘Driver confusion status detection using recurrent
IFAC-PapersOnLine, vol. 51, no. 9, pp. 267–272, Jan. 2018, doi: neural networks,’’ in Proc. IEEE Int. Conf. Multimedia Expo (ICME),
10.1016/j.ifacol.2018.07.044. Seattle, WA, USA, Jul. 2016, pp. 1–6, doi: 10.1109/ICME.2016.7552966.
[88] F. Zhang, S. Mehrotra, and S. C. Roberts, ‘‘Driving distracted [106] Y. Feng, S. Pickering, E. Chappell, P. Iravani, and C. Brace, ‘‘Driving style
with friends: Effect of passengers and driver distraction on young analysis by classifying real-world data with support vector clustering,’’
drivers’ behavior,’’ Accident Anal. Prevention, vol. 132, Nov. 2019, in Proc. 3rd IEEE Int. Conf. Intell. Transp. Eng. (ICITE), Singapore,
Art. no. 105246, doi: 10.1016/j.aap.2019.07.022. Sep. 2018, pp. 264–268, doi: 10.1109/ICITE.2018.8492700.
[89] K. Nakano and B. Chakraborty, ‘‘Effective feature extraction from driving [107] D. S. Lee, T. W. Chong, and B. G. Lee, ‘‘Stress events detection of driver
data for detection of danger awareness,’’ in Proc. IEEE 10th Int. Conf. by wearable glove system,’’ IEEE Sensors J., vol. 17, no. 1, pp. 194–204,
Awareness Sci. Technol. (iCAST), Morioka, Japan, Oct. 2019, pp. 1–6, Jan. 2017, doi: 10.1109/JSEN.2016.2625323.
doi: 10.1109/ICAwST.2019.8923343. [108] L. Precht, A. Keinath, and J. F. Krems, ‘‘Effects of driving anger
[90] M. H. Alkinani, W. Z. Khan, and Q. Arshad, ‘‘Detecting human driver on driver behavior—Results from naturalistic driving data,’’ Transp.
inattentive and aggressive driving Behavior using deep learning: Recent Res. F, Traffic Psychol. Behav., vol. 45, pp. 75–92, Feb. 2017, doi:
advances, requirements and open challenges,’’ IEEE Access, vol. 8, 10.1016/j.trf.2016.10.019.
pp. 105008–105030, 2020, doi: 10.1109/ACCESS.2020.2999829. [109] S. Jomnonkwao, S. Uttra, and V. Ratanavaraha, ‘‘Analysis of a driving
[91] M. S. Wang, N. T. Jeong, K. S. Kim, S. B. Choi, S. M. Yang, S. H. You, behavior measurement model using a modified driver behavior ques-
J. H. Lee, and M. W. Suh, ‘‘Drowsy behavior detection based on driving tionnaire encompassing texting, social media use, and drug and alcohol
information,’’ Int. J. Automot. Technol., vol. 17, no. 1, pp. 165–173, consumption,’’ Transp. Res. Interdiscipl. Perspect., vol. 9, Mar. 2021,
Feb. 2016, doi: 10.1007/s12239-016-0016-y. Art. no. 100302, doi: 10.1016/j.trip.2021.100302.
[92] A. D. McDonald, J. D. Lee, C. Schwarz, and T. L. Brown, ‘‘A con- [110] Y. Üzümcüoğlu and T. Özkan, ‘‘Traffic climate and driver behav-
textual and temporal algorithm for driver drowsiness detection,’’ iors: Explicit and implicit measures,’’ Transp. Res. F, Traffic Psy-
Accident Anal., Prevention, vol. 113, pp. 25–37, Apr. 2018, doi: chol. Behav., vol. 62, pp. 805–818, Apr. 2019, doi: 10.1016/j.trf.
10.1016/j.aap.2018.01.005. 2019.03.016.
[93] C. J. De Naurois, C. Bourdin, A. Stratulat, E. Diaz, and J. L. Vercher, [111] Y. Üzümcüoğlu, T. Özkan, C. Wu, and H. Zhang, ‘‘Traffic cli-
‘‘Detection and prediction of driver drowsiness using artificial neu- mate and driver behaviors: The moderating role of driving skills in
ral network models,’’ Accident Anal. Prevention, vol. 126, pp. 95–104, Turkey and China,’’ J. Saf. Res., vol. 75, pp. 87–98, Dec. 2020, doi:
May 2019, doi: 10.1016/j.aap.2017.11.038. 10.1016/j.jsr.2020.08.004.
[94] M. Shahverdy, M. Fathy, R. Berangi, and M. Sabokrou, ‘‘Driver behav- [112] E. Zicat, J. M. Bennett, E. Chekaluk, and J. Batchelor, ‘‘Cognitive func-
ior detection and classification using deep convolutional neural net- tion and young drivers: The relationship between driving, attitudes, per-
works,’’ Expert Syst. Appl., vol. 149, Jul. 2020, Art. no. 113240, doi: sonality and cognition,’’ Transp. Res. F, Traffic Psychol. Behav., vol. 55,
10.1016/j.eswa.2020.113240. pp. 341–352, May 2018, doi: 10.1016/j.trf.2018.03.013.
[95] K. Saleh, M. Hossny, and S. Nahavandi, ‘‘Driving behavior classification [113] J. Wu and H. Xu, ‘‘Driver behavior analysis for right-turn drivers
based on sensor data fusion using LSTM recurrent neural networks,’’ in at signalized intersections using SHRP 2 naturalistic driving study
Proc. IEEE 20th Int. Conf. Intell. Transp. Syst. (ITSC), Yokohama, Japan, data,’’ J. Saf. Res., vol. 63, pp. 177–185, Dec. 2017, doi: 10.1016/j.jsr.
Oct. 2017, pp. 1–6, doi: 10.1109/ITSC.2017.8317835. 2017.10.010.
[96] M. A. Khodairy and G. Abosamra, ‘‘Driving behavior classification [114] A. Calvi, ‘‘Investigating the effectiveness of perceptual treatments
based on oversampled signals of smartphone embedded sensors using on a crest vertical curve: A driving simulator study,’’ Transp. Res.
an optimized stacked-LSTM neural networks,’’ IEEE Access, vol. 9, F, Traffic Psychol. Behav., vol. 58, pp. 1074–1086, Oct. 2018, doi:
pp. 4957–4972, 2021, doi: 10.1109/ACCESS.2020.3048915. 10.1016/j.trf.2018.06.002.

14152 VOLUME 11, 2023

S. Bouhsissin et al.: Driver Behavior Classification: A Systematic Literature Review

[115] K. Chang, M. V. Ramirez, B. Dyre, M. Mohamed, and A. Abdel-Rahim, [134] T. A. Dingus, S. Klauer, V. R. Lewis, A. Petersen, and S. E. Lee,
‘‘Effects of longitudinal pavement edgeline condition on driver lane ‘‘The 100-car naturalistic driving study phase II—Results of the 100-car
deviation,’’ Accident Anal. Prevention, vol. 128, pp. 87–93, Jul. 2019, doi: field experiment,’’ U.S. Dept. Transp., Tech. Rep. DTNH22-00-C-07007,
10.1016/j.aap.2019.03.011. 2006.
[116] J. Rogé, S. Laurent, D. Ndiaye, I. Aillerie, and F. Vienne, ‘‘Does a [135] V. Neale, T. Dingus, S. Klauer, and J. Sudweeks, ‘‘An overview
yellow jacket enhance cyclists’ sensory conspicuity for car drivers during of the 100-car naturalistic study and findings,’’ in Proc. Int. Tech.
daylight hours in an urban environment?’’ Saf. Sci., vol. 119, pp. 385–391, Conf. Enhanced Saf. Vehicles, vol. 19, Jan. 2005. [Online]. Available:
Nov. 2019, doi: 10.1016/j.ssci.2018.06.023. https://www.researchgate.net/publication/237252699
[117] J. Klieger and I. Savage, ‘‘Motor-vehicle drivers’ behavioral response to [136] H. Ouchra and A. Belangour, ‘‘Object detection approaches in images:
increased bicycle traffic,’’ J. Saf. Res., vol. 74, pp. 97–102, Sep. 2020, A survey,’’ in Proc. 13th Int. Conf. Digit. Image Process. (ICDIP),
doi: 10.1016/j.jsr.2020.04.002. Jun. 2021, p. 85, doi: 10.1117/12.2601452.
[118] A. Das, A. Ghasemzadeh, and M. M. Ahmed, ‘‘Analyzing the effect of fog
weather conditions on driver lane-keeping performance using the SHRP2
naturalistic driving study data,’’ J. Saf. Res., vol. 68, pp. 71–80, Feb. 2019,
doi: 10.1016/j.jsr.2018.12.015.
[119] K. Yoneda, A. Kuramoto, and N. Suganuma, ‘‘Convolutional neural
network based vehicle turn signal recognition,’’ in Proc. Int. Conf.
Intell. Informat. Biomed. Sci. (ICIIBMS), Nov. 2017, pp. 204–205, doi:
10.1109/ICIIBMS.2017.8279693.
[120] S. J. Kamble and M. R. Kounte, ‘‘Machine learning approach on SOUKAINA BOUHSISSIN received the B.Sc.
traffic congestion monitoring system in Internet of Vehicles,’’ degree in mathematical sciences and computer sci-
Proc. Comput. Sci., vol. 171, pp. 2235–2241, Jan. 2020, doi: ence from the Faculty of Sciences Ben M’Sick,
10.1016/j.procs.2020.04.241. Hassan II University of Casablanca, Morocco,
[121] A. Sheykhfard, F. Haghighi, E. Papadimitriou, and P. Van Gelder, ‘‘Anal- in 2018, and the M.Sc. degree in data science and
ysis of the occurrence and severity of vehicle-pedestrian conflicts in big data from Hassan II University of Casablanca,
marked and unmarked crosswalks through naturalistic driving study,’’ in 2020, where she is currently pursuing the Ph.D.
Transp. Res. F, Traffic Psychol. Behav., vol. 76, pp. 178–192, Jan. 2021, degree in computer science. Her research interests
doi: 10.1016/j.trf.2020.11.008. include driver behavior classification, intelligent
[122] Y. Xiao and M. Watson, ‘‘Guidance on conducting a systematic literature transport systems, machine learning, deep learn-
review,’’ J. Planning Educ. Res., vol. 39, no. 1, pp. 93–112, Aug. 2019,
ing, image classification, and temporal series.
doi: 10.1177/0739456X17723971.
[123] B. Kitchenham and S. M. Charters, ‘‘Guidelines for performing system-
atic literature reviews in software engineering,’’ Keele Univ., Durham
Univ., Joint Report EBSE 2007-001, 2007.
[124] L. M. Martinussen, L. Hakamies-Blomqvist, M. Møller, T. Özkan, and
T. Lajunen, ‘‘Age, gender, mileage and the DBQ: The validity of the driver
behavior questionnaire in different driver groups,’’ Accident Anal. Pre-
vention, vol. 52, pp. 228–236, Mar. 2013, doi: 10.1016/j.aap.2012.12.036.
[125] T. Özkan, T. Lajunen, and H. Summala, ‘‘Driver behaviour question- NAWAL SAEL received the engineering degree
naire: A follow-up study,’’ Accident Anal. Prevention, vol. 38, no. 2, in software engineering from ENSIAS, Morocco,
pp. 386–395, Mar. 2006, doi: 10.1016/j.aap.2005.10.012. in 2002. She has been a Teacher-Researcher, since
[126] I. van Schagen and F. Sagberg, ‘‘The potential benefits of naturalistic driv- 2012, an Authorized Professor, since 2014, and a
ing for road safety research: Theoretical and empirical considerations and Professor of higher education with the Department
challenges for the future,’’ Proc. Social Behav. Sci., vol. 48, pp. 692–701,
of Mathematics and Computer Science, Faculty
Jan. 2012, doi: 10.1016/j.sbspro.2012.06.1047.
of Sciences Ben M’Sick, Hassan II University of
[127] J. Bärgman, ‘‘Methods for analysis of naturalistic driving data in driver
Casablanca, Casablanca, since 2020. Her research
behavior research: From crash-causation analysis using expert assessment
to quantitative assessment of the effect of driver behavior on safety using
interests include data mining, educational data
counterfactual simulation,’’ M.S. thesis, Dept. Appl. Mech., Chalmers mining, machine learning, deep learning, and the
Tekniska Högskola, Gothenburg, Sweden, 2016. Internet of Things.
[128] K. Kircher and C. Ahlström, ‘‘Can we trust driver behaviour assessment?
Examples from research in simulators and in the field,’’ in Proc. Measur-
ing Behav., 2012, pp. 55–58.
[129] G. S. Larue, C. Wullems, M. Sheldrake, and A. Rakotonirainy, ‘‘Valida-
tion of a driving simulator study on driver behavior at passive rail level
crossings,’’ Hum. Factors, J. Hum. Factors Ergonom. Soc., vol. 60, no. 6,
pp. 743–754, Sep. 2018, doi: 10.1177/0018720818783507.
[130] SHRP2 NDS Data Access. Accessed: Dec. 22, 2022. [Online]. Available:
FAOUZIA BENABBOU has been a Teacher-
https://insight.shrp2nds.us/
Researcher, since 1994, an Authorized Professor,
[131] E. Romera, L. M. Bergasa, and R. Arroyo, ‘‘Need data for driver
since 2008, and a Professor of higher education
behaviour analysis? Presenting the public UAH-DriveSet,’’ in Proc. IEEE
with the Department of Mathematics and Com-
19th Int. Conf. Intell. Transp. Syst. (ITSC), Rio de Janeiro, Brazil,
Nov. 2016, pp. 387–392, doi: 10.1109/ITSC.2016.7795584. puter Science, Ben M’Sick Faculty of Sciences,
[132] T. Seo et al., ‘‘Evaluation of large-scale complete vehicle trajectories Hassan II University of Casablanca, Casablanca,
dataset on two kilometers highway segment for one hour duration: Zen since 2015. She is a member of the Informa-
Traffic Data,’’ U.S. Dept. Transp. Intell. Transp. Syst. Joint Program tion Technology and Modeling Laboratory and the
Office (JPO), USA, 2016, doi: 10.21949/1504477. Leader of the Cloud Computing, Network and Sys-
[133] D. Bezzina and J. Sayer, ‘‘Safety pilot model deployment: Test con- tems Engineering (ICCNSE) Team. Her research
ductor team report,’’ Nat. Highway Traffic Saf. Admin., Washington, interests include cloud computing, data mining, machine learning, and natu-
DC, USA, Tech. Rep. DOT HS 812 171, 2015. [Online]. Available: ral language processing.
https://www.ntis.gov

VOLUME 11, 2023 14153

View publication stats