A Comparative Study on Traffic Modeling Techniques for Predicting and Simulating

Traffic Behavior Alghamdi

Models study the spatial and temporal interactions within the traffic status to capture their
significant effect on traffic behavior. The temporal factor in traffic models is used to identify a
directional movement in a time series format, while the spatial factor will take into account the
geographical space to build the prediction outcome. By combining the temporal and spatial factors
in traffic prediction models, an improvement in the prediction results is noticeable [4]. Numerous
studies have proven that machine learning models are capable of modeling the spatiotemporal
correlations in continuous spatiotemporal traffic data.

this study show that the prediction accuracy increases when the targeted time window is more
than 10 min, while the prediction of a time window that is less than 10 min suffers from instability.

Although studies claimed that long-term traffic prediction is not beneficial to obtain an accurate
prediction, other studies that apply traffic modeling for long-term prediction highlighted the
importance of long-term prediction to improve traffic management systems [16]. Because of the
large time window in the long-term prediction, seasonal patterns and cycle patterns will be
detected in the traffic data. Another study indicates that exponential smoothing models are
powerful in capturing seasonality in the traffic data as well as handling trends and white noise
satisfactorily.

There are a number of challenges in the field of traffic prediction modeling concerning time
intervals. Employing the traffic prediction modeling for long-term time intervals faces several
issues. aggregating traffic data will lead to a high rate of errors in the prediction outcome. On the
other hand, modeling short-term traffic prediction requires high computations and is highly
sensitive to outliers. Therefore, using data analysis in traffic modeling plays an important role in
reducing the drawbacks of predicting short-term or long-term traffic status, where it provides
tools and functions that help in cleansing and transforming data into useful information before
applying prediction models.

A Comprehensive Review on Traffic Prediction for Intelligent Transport System Suhas

A Survey on Modern Deep Neural Network for Traffic Prediction: Trends, Methods and
Challenges Tedjopurnomo

Traffic prediction is different from conventional time-series analysis in that traffic prediction is
subject to spatial factors as well as many other external factors. For instance, the prediction of
traffic at one site depends on the traffic at other sites and all of the sites are affected by external
factors such as weather and holidays. Amongst all the available traffic prediction methods, deep
neural network is the most prominent. This is due to its sheer predictive power that can model the
complex and nonlinear traffic patterns [4], [5], [6], [7], [8]. The three most common deep neural
network models used for traffic prediction are Convolutional Neural Networks, Recurrent Neural
Networks, and Feedforward Neural Networks.

Deep neural networks consist of complex neural network models with a large number of layers.
Some examples are Recurrent Neural Network, Convolutional Neural Network, Feedforward
Neural Network, and hybrids of these models. Some of the deep neural network models can
explicitly capture different aspects of traffic data, which made them even more attractive. For
instance, CNN can explicitly capture the spatial aspect of traffic data while RNN can explicitly
capture the temporal aspect of traffic data. Additionally, the increased number of layers improves
the models’ prediction capability. This factor allows them to model traffic fluctuations more
accurately.

While the strengths of deep neural network models made them attractive, they also possess
several disadvantages compared to the older methods:

 Deep Neural Network Models Require a Large Amount of Data That Covers all Traffic
Conditions. If the amount of data is too small or if the data is not diverse enough, the
model’s generalization capability is compromised.
 Deep Neural Network Models Still take a Long Time to Train. As deep neural network
models are complex and have a large number of layers, the training time can be very long.
 This problem is compounded on hybrid deep neural network models. As classical statistical
and older machine learning models are not as complex, their training time is much shorter.
 Deep Neural Network Models are Difficult to Interpret. This is because of two reasons: the
number of internal parameters is very large, and the parameters are learned from training,
not set manually. Thus, while they can predict well, it is hard to understand their
parameters. Understanding the parameters may reveal important information such as the
spatiotemporal dynamics in the road network.

Afterwards, machine learning models improve upon the performance of classical statistical
models. Then, the deep neural network class of models dominates the field due to its capability in
capturing the complex and nonlinear patterns in traffic data.

Data Time Range. One major deficiency we have observed in the field of deep neural network for
traffic flow prediction is the insufficient data time range. 26 out of 37 literatures in Tables 4 and 5
use less than one year’s worth of data. This deficiency will have an adverse impact on subtropical
regions, as seasonal changes may affect temperature and weather, which in turn can affect traffic.
By using data from only one or a few months, the model cannot generalize to different seasons.
This can be mitigated by incorporating weather data, but as mentioned before, this is a difficult
and time-consuming task. Some authors also use data from only a certain range of hours or use
data from weekdays only. This will also cause problems as the model cannot generalize well to
situations outside the boundaries of the provided data. For instance, using traffic data from 07.00
AM to 11.00 PM only may reduce the model’s performance on the excluded hours, and using only
weekdays data may adversely impact the model’s performance when predicting weekend traffic.
Deep neural network models are flexible and can adapt well to data. Consequently, using vastly
different datasets may result in an entirely different model. Thereby, for a model to be applicable
to real application scenarios, it is important to use a dataset that closely resembles those
scenarios.

PROBLEM DEFINITION Within this attribute, we have identified four sublevels. The first sublevel is
related to the type of data used as an input. Based on [116], there are three types of inputs that
can be used at the moment of feeding a ML model:

1. only using temporal traffic data,

2. temporal and spatial traffic data,
3. and non-traffic data (e.g. calendar data) to enrich any of the two aforementioned inputs.

First of all, it is significant to understand what the traffic prediction problem means. Therefore, we
will use some examples to show the concept of traffic prediction:

– Traffic status prediction: It is popular to use the navigation system of the electronic map to avoid
congested roads when we plan to leave one place for another. The key ability to achieve the target
is to predict which roads will be congested in the future time. In other words, we need to predict
the traffic status for each road. However, it is typical to measure traffic status with average traffic
speed or travel time. The slower the traffic speed or the more the travel time, the worse the traffic
status. Therefore, the traffic status prediction can be regarded as the traffic speed or travel time
prediction, which are regression problems. Moreover, we can measure the traffic status with
different types (e.g., smooth, light congestion and heavy congestion) by splitting the traffic speed
into different continuous intervals, where predicting the traffic status becomes a classification
problem.

– traffic flow prediction: Recently, there exist some stomp events caused by excessive traffic. The
main reason is that the government cannot monitor and guide the flow of people in time. Hence, it
is significant to predict traffic flows in future time. Moreover, traffic flow can be divided into two
types: network-based and region-based. The first type infers the number of vehicles collected by
loop detector sensors, which are installed on both endpoints of the roads. As for the second type,
we split the whole city into different regions and regard the number of crowds leaving one region
for another as the region-based traffic flow. Therefore, the region-based traffic flow can be further
divided into in-flow and out-flow. For example, if there are 100 people leaving the region A for the
region B, both A’s out-flow and B’s in-flow would increase 100.

– Travel demand prediction: Transportation companies provide online taxi service for users. They
need to predict people’s travel demands in order to better dispatch vehicles for different regions.
For example, they should dispatch more vehicles to residential areas during the morning rush
hour. In contrast, they should dispatch more vehicles to office zones during the evening rush hour.
Generally, predicting travel demands is based on regions, so we also call it region-based travel
demand prediction.

Spatio-temporal data. As mentioned before, the trafc prediction is based on collected spatio-
temporal data, such as road network and historical trafc data. Actually, the spatiotemporal data,
related to trafc prediction problems, include road network, region-based trafc (e.g., regional fows
and travel demands), network-based trafc (e.g., intersection fows, road speed and travel time),
trajectory, POI features, event data, meteorological data and temporal data (holiday, date and
timestamp). As shown in Table 1, inspired by [1, 2], the above spatio-temporal data can be broadly
categorized into the following fve categories:

(1) SO (Spatial-Only data). The data only include spatial features and have no temporal features,
such as the POI information and the road network in a city.

(2) TO (Temporal-Only data). The data only include temporal features and have no spatial features,
such as the date, the timestamp and the holiday.

(3) STS (Spatio-temporal Static data). For this kind of data, there is no change in both the spatial
dimension and the temporal dimension, such as event data.

(4) SSTD (Spatial Static Temporal Dynamic data). The data can be regarded as a sequence in the
temporal dimension and keep static in the spatial dimension. The components include trafc fows,
travel demands, travel time, trafc speed and meteorological data (i.e., weather).
(5) SDTD (Spatial Dynamic Temporal Dynamic data). The data can be regarded as a sequence in
both the spatial and temporal dimensions, such as trajectories. We will discuss more details in
Sect. 2.

The accuracy of predictions can be influenced by factors such as the distribution of road
networks, as well as weather conditions, accidents, holidays, and other similar variables.
When making predictions for an entire city, it is important to take into account the intricate
topological connections of the road network within the metropolitan area. Moreover, city-
wide forecasting is a common task in time series prediction, and accurately capturing the
long-term temporal relationship between past observations is challenging for traditional
forecasting methods, particularly when accounting for periods and trends. In addition,
environmental circumstances such as heavy rainfall, traffic accidents, significant festivals,
and so on, have an influence on urban traffic conditions.
Although statistical methods have been widely adopted, they cannot effectively extract inherent
spatial and temporal dependencies and remain limited to a single road segment scenario. To
predict city-wide traffic flow volume, a large number of independent models must be constructed.
This limits the further application of statistical methods in the city-wide traffic flow prediction
problema.

ANNs have not been widely used because of their poor generalization ability caused by their
inherently shallow architecture. At present, several advanced and powerful deep learning models
have been introduced into the academic community to solve traffic flow prediction problems. For
example, the work in [16] utilized deep learning methods to predict the traffic status of a local
region. The researchers in [17] were the first to adopt Deep Belief Networks (DBNs) into the traffic
forecasting field and achieved favourable results. The study in [18] introduced an RBM-RNN model
that combined deep restricted Boltzmann machines (RBMs) with a recurrent neural network
(RNN), which inherited the advantages of the above two models. The work in [19] constructed a
deep-learning-based prediction model and utilized a stack auto-encoder (SAE) to extract local
spatiotemporal features. Deep learning models have achieved better results than traditional
methods because they possess much deeper and more complicated structures. However, few
studies have involved city-wide traffic prediction problems and especially lack enough
consideration of both spatiotemporal dependencies and the complex external influences hidden in
the traffic system.