CHAPTER 4

Applications of neuro fuzzy systems: A brief review and future outline

Under this chapter we present neuro fuzzy systems (NFS) development using classification and
literature review of articles for the last decade (2002–2012) to explore how various NFS
methodologies have been developed during this period. Based on the selected journals of
different NFS applications and different online database of NFS, this article surveys and
classifies NFS applications into ten different categories such as student modeling system,
medical system, economic system, electrical and electronics system, traffic control, image
processing and feature extraction, manufacturing and system modeling, forecasting and
predictions, NFS enhancements and social sciences. Our study indicates mainly three types of
future development directions for NFS methodologies, domains and article types: (1) NFS
methodologies are tending to be developed toward expertise orientation. (2) It is suggested that
different social science methodologies could be implemented using NFS as another kind of
expert methodology. (3) The ability to continually change and learning capability is the driving
power of NFS methodologies and will be the key for future intelligent applications.

4.1. Introduction

Applications of artificial intelligence (AI) in various fields are getting more and more popular
during the last decade and that is why much relevant research has been conducted. AI methods
are mainly comprised of fuzzy logic, neural networks, genetic programming and hybrid
approaches such as neuro fuzzy systems, genetic fuzzy systems and genetic programming neural
networks etc. Neuro-fuzzy systems refer to combinations of artificial neural lnetwork and fuzzy
logic in the field of artificial intelligence, which was proposed by Jang [1] in 1993. The basic
idea behind this NFS is that it combines human-like reasoning style of fuzzy systems with the
learning and connectionist structure of neural networks. NFS provides powerful and flexible
universal approximations with the ability to explore interpretable IF-THEN rules. The use of
NFS is proliferating into many sectors in our social and technological life. As a part of NFS
research, this paper surveys the development of NFS through a literature review and
classification of articles from 2002 to 2012 based on applied methodologies, domains and article
types. Based on the survey, this paper explores a future research direction of NFS in various
domains. The reason behind choosing this period is that many authors started to shift to neuro
fuzzy domains for much improved result and a huge number of related applications were
developed in this duration. This literature survey is based on a search in the keyword index and
article abstract for ‘NFS’ on the IEEE Xplore, Elsevier, ACM digital library, Springer, Ingenta,
EBSCO, Wiley Inter Science online database and many other reputed journals and conference
proceedings. Based on the scope of collected articles on NFS applications, this paper surveys and
classifies NFS applications into ten different categories: student modeling system, medical
system, economic system, electrical & electronics system, traffic control, image processing &
feature extraction, manufacturing & system modeling, forecasting & predictions, NFS
enhancements and social sciences for different research and problem domains.

2. NFS in student modeling

Neuro fuzzy applications in educational field are getting more and more popular during the last
decade (2002–2012) and that is why much relevant research has been conducted. Neuro fuzzy
system has a wide range of applications in the educational field and new directions are constantly
given in educational research. According to Stathakopoulou et al. [2] student modeling is
consisted of two components: the student model and the diagnostic module. The student model is
one of the components of an intelligent tutoring system (ITS) which provides a description of
student related information such as his knowledge level, skills or even preferences while
diagnosis is the inference process which results in the end updates of the student model. Student
modeling includes student classification, monitoring students’ actions, processing intelligent
learning environment (ILE), assessing students’ knowledge, evaluating students in intelligent
tutoring system, modeling students in web based ITS etc. A summary of thoughts contributed by
various authors in this category is described below. During the last decade a number of
researchers have contributed their innovations in this category. In the year 2003, Stathacopoulou
et al. [3] presented a neural network-based fuzzy modeling approach to assess student
knowledge. Authors used fuzzy logic to handle the subjective judgments of human tutors with
respect to student observable behavior and knowledge, here Student knowledge is decomposed
into pieces and assessed by combining fuzzy evidences, each one contributing to some degree to
the final assessment. Later in 2004, they implemented a neural network and fuzzy logic based
model to achieve accurate student diagnosis and updates the student model in intelligent learning
system (ILS) [4]. In the same year, they presented neural network-based fuzzy model [5] to
access student’s motivational state in a discovery learning environment with the assistance of a
group of expert teachers. In 2004, Ali et al. [6] presented a neural fuzzy inference system for
modeling student in the context of web based intelligent tutoring system. Web based intelligent
tutoring system provides a learner platform independence, time independence and location
independence and also without the expense of one to one tutoring scopes. Later in 2006, Sevarac
[7] presented a neuro-fuzzy system for student modeling. The proposed system performed
classification of students based on qualitative observations of their characteristics. Next year
Stathacopoulou et al. [8] implemented a neural network based fuzzy diagnostic model by
monitoring students’ learning style and teachers’ expertise behavior. Authors used fuzzy logic to
provide a linguistic description of students’ behavior and learning characteristics. They used
neural networks to add learning and generalization abilities to the fuzzy model by encoding
teachers’ experience through supervised neural-network learning. In 2009, Taylan et al. [9]
proposed adaptive neuro fuzzy inference system (ANFIS) using genetic algorithm (GA) to assess
student’s academic performance. Recently, Iraji et al. [10] proposed student classification
technique using adaptive neuro fuzzy system. Adaptive neuro fuzzy system has established itself
as one of the popular modeling technique in the field of control systems, expert systems and
complex system modeling. Their motivation behind the research work was to identify
exceptional and week students for proper attention. The methodology, domain and article type of
stu-dent modeling system are categorized in Table 4.1.

Figure 4.1 Student modeling system

3. NFS in medical system

A medical system, also sometimes referred to as health care system is the organization of people,
institutions and resources to deliver health care services to meet the health needs of target
populations. Presently diseases in India have emerged as number one killer in both urban and
rural areas of the country. It will be of greater value if the diseases are diagnosed in its early
stage. Correct diagnosis of the disease will decrease the death rate due to different diseases.
Many clinical tests are being done to find the presence of the disease. In last decade neuro fuzzy
applications in medical system are getting huge attention and that is why much relevant research
has been conducted. NFS are being used for various typical disease diagnoses like brain disorder,
cardiac disease, breast cancer, alzheimer, thyroid disorder, leukemia, hypotension, heart disease
etc. Following is a brief outline that shows how neuro fuzzy system has been contributing in
medical disease diagnosis since 2003.In 2003, Neagoe et al. [12] proposed fuzzy gaussion
neural network to recognize the ECG signals for ischemic heart disease diagnosis with the help
of applied Principal Component Analysis(PCA) and Discrete Cosine Transform (DCT).
Mastorocostas and Hilas [13] in 2004 presented fuzzy neural filter for the separation of
discontinuous lung sounds using dynamic orthogonal least square method. Later in 2005
Mastorocostas and Theocharis [14] presented the same fuzzy neural filter for real time separation
of discontinuous sounds from lung obtained from the patients. In the same year Guler et al. [15]
proposed adaptive neuro fuzzy inference system (ANFIS) model for classification of electro
encephalogram (EEG) signals, which was clinically used to investigate brain disorder. Oweis et
al. [16] in 2005 proved that neuro-fuzzy approach could be one of the important modalities in
image processing, especially in the biomedical scenario where most diagnostic techniques rely
on imaging. The presented pixel classification tool was a highly useful, powerful, and
independent of any expert or any prior knowledge. After a couple of years Stavrakoudis et al.
[17] presented the fuzzy neural filter to perform separation of lung sounds involving the patients
with pulmonary pathology. They used Takagi–Sugeno–Kang (TSK) recurrent network for this
purpose. In the same year (2007), Polat and Gunes [18] paid their attention for diabetes patients.

Table 4.2. Medical system

Using expert system, Principal Component Analysis (PCA) and proposed ANFIS, they started to
improve diagnostic accuracy of diabetes disease. They reached to the classification accuracy
point which was 89.47%, promising with regard to other classification applications. Sengur [19]
in 2008investigated the use of linear discriminant analysis (LDA) and ANFIS to determine the
normal and abnormal heart valves by diagnosing heart sounds. Ubeyli [20] in 2009 presented
ANFIS for ECG signal classification using Lyapunov exponents for improvement of diagnostic
accuracy. Later in 2010, Ovreiu and Simon [21] focused on cardiac disease. They used
biogeography based optimization(BBO), Evolutionary Algorithm (EA), neuro fuzzy rules for
cardiac diagnosis. In the same year, Alamelumangai and DeviShree [22] proposed particle
swarm optimization (PSO) algorithm to optimize neuro fuzzy model for ultrasound image
segmentation. This system was tested for detecting the micro calcifications in breast sonograms.
Obi and Imainvan [23] in 2011 analyzed Alzheimer disease diagnosis using neuro fuzzy
inference procedure. In that year they also paid attention on lukemia diagnosis [24]. For
diagnosis they used neuro fuzzy logic (NFL) and expert system based procedure. Kumar et al.
[25] in 2011 proposed fast ANFIS to detect the occurrence of cancer. Fast adaptive neuro fuzzy
inference sys-tem (FANFIS) was used along with gene ranking technique called analysis of
variance (ANOVA). The learning technique used in this paper is modified Levenberg–Marquardt
algorithm. Recently in2012, Agboizebeta and Chukwuyeni [26] have focused on thyroid disorder
with the help of neuro fuzzy expert system using a set of symptoms. The system designed was an
interactive system that was able to interact with the patient briefing his current condition. They
also proposed cognitive neuro fuzzy expert system for hypotension control [27] in the same year.
Ephzibah and Sundarapandian [28] concentrated on medical system for diagnosis of heart
disease. They used combination of neuro fuzzy technique and other soft computing techniques
like genetic algorithm. Khamenehet al. [29] in 2012 has performed investigation using ANFIS to
detect abnormality in red blood cell and classify blood samples into normal and abnormal
category. Their motivation behind the research work was to identify, classify, diagnosis different
type of disease for different fields. The methodology, domain and article type of medical system
are categorized in Table 2.

4. NFS in economic system

An economic system can be defined as an organization where a person, country or area makes,
distributes, consumes, buys or sells services and goods. This type of system has a direct impact
on various governments and also on public activities. NFS can be applied in various field of
economic system like state economic, stock market, toll collection, gas condensate, energy
consumption, electric load forecasting, price prediction, supply chain management etc. Since the
last decade, Neuro fuzzy applications in economic systems are attaining huge attention of many
researchers and a number of relevant researches have been conducted. Starting from stock
market to supply chain network, NFS has a wide range of applications in the economic systems,
a few of those are briefly described here. In the year 2005, Ang and Quek [31] proposed rough
set-based pseudo outer-product (RSPOP) for stock market prediction. The proposed stock trading
model avoids the forecast bottleneck and synergizes the time delayed price difference forecast
approach with simple moving average rules for generating trading signals. LaterLin et al. [32]
provided a mathematical model using neuro fuzzy approach to describe the online English
auction process for final price prediction. In this paper, authors proposed a hybrid neuro fuzzy
method to predict the final price in addition to exploring the complicated possibly nonlinear
relationship among the auction mechanisms and final price. The empirical results showed that
neuro fuzzy system can predict the final price accurately much better than the others. Kablan
[33] in 2009 focused on financial trading market. Using ANFIS model author achieved better
performance in trading system. In the same year, Nowroozi et al. [34] proposed NFS for gas
condensate. Gumus et al. [35] did work on supply chain network using neuro fuzzy procedure
and mixed integer linear programming (MILP). Atsalakis and Valavanis [36] also directed their
research on stock market using ANFIS. In 2010, Mord-jaoui et al. [37] presented a short term
electric load forecasting model using ANFIS for forecasting electric load in industry. Iranmanesh
et al. [38] in 2011 proposed a forecast method based on local linear neuro fuzzy (LLNF) model
and fuzzy transform mechanism for energy consumption. Schott and Kalita [39] also presented
ANFIS in stock market prediction. Recently Kasa [40] used neuro fuzzy system and fuzzy logic
for economical decision making. In the same year, Fang [41] proposed ANFIS in the application
of financial crisis forecasting. Tsai and Chen [174] explored various theories related to different
kinds of natural disaster risk analysis mechanisms, with the goal of establishing a rapid risk

Table 4.3. Economic system

assessment model suited to the tourism industry that can be used to quickly analyze disaster-
forming characteristics and risk weaknesses in local regions. Last year Giovanis [42] predicted
economic crisis period in USA using two approaches: the first approach includes Logit and
Probit models and the second is an adaptive neuro-fuzzy inference system with Gaussian and
generalized bell membership functions. Liu et al. [157] combined three categories of research
(risk assessment based, life cycle assessment based and criterion based) in an integrated
framework to retrieve the benefits of these three combined categories. Lin et al. [154] aimed to
develop a model for debris flow hazard assessment in Taiwanusing Kalman filter decision
system. Besides these, a number of researchers [141,142,145–148,150,151,153,163–
166,168,173]have contributed in this category. Researchers’ motivation behind this research
work was to predict and solve economic crisis of various field. The methodology, domain and
article type of economic system are categorized in Table 4.3. NFS in traffic control Neuro fuzzy
system has a wide range of applications in the traffic control since last decade. Recently a
number of researchers are paying their attention in this category. Road traffic control is the
process which is used to describe how councils and highway authorities control use of the road
network in order to achieve improvements in road safety and efficiency. Network traffic control
is the process of managing, prioritizing, controlling or reducing the network traffic to reduce
congestion, latency and packet loss. During the last decade a number of researchers have given
their concentration in this category. In 2005 Abdennour [44] developed neuro fuzzy system for
MPEG coded video traffic control which is important for networking and internet multimedia.
The reason behind this development of the system was to simplify the sys-tem design and also
for fast execution. The author proved that neuro fuzzy systems could be very effective in long-
term prediction of this type of traffic. Izquierdo et al. [45] in 2007, proposed neuro fuzzy
approaches for dynamic classification of the road vehicles to avoid road vehicles collision. In the
same year, Partoucheet al. [46] presented a new approach to autonomously adapt the speed of a
vehicle by learning from a human driver and using anticipation. Later, Aseri and Bagai [47]
proposed ANFIS to solve the problem of rate control for Available Bit Rate (ABR) service class
in Asynchronous Transfer Mode (ATM) networks. Sindal et al. [48]in 2009 developed neuro
fuzzy call admission control algorithm to reduce voice data traffic in CDMA cellular network by
trans-mitting signals as lower power level. In 2011, Abbas et al. [49] proposed neuro fuzzy
system for best route selection to avoid traffic congestion. They used neuro fuzzy logic and ant
colony system (ACS) algorithm for routing. The researcher’s motivation behind the research
work was to avoid road traffic, road accident, network congestion, video traffic and vehicles
congestion. The methodology, domain and article type of traffic control categorized in Table 4.3.

5. NFS in traffic control

Neuro fuzzy system has a wide range of applications in the traffic control since last decade.
Recently a number of researchers are paying their attention in this category. Road traffic control
isthe process which is used to describe how councils and highway authorities control use of the
road network in order to achieve improvements in road safety and efficiency. Network traffic
control is the process of managing, prioritizing, controlling or reducing the network traffic to
reduce congestion, latency and packet loss. During the last decade a number of researchers have
given their concentration in this category. In 2005 Abdennour [44] developed neuro fuzzy
system for MPEG coded video traffic control which is important for networking and internet
multimedia. The reason nbehind this development of the system was to simplify the system
design and also for fast execution. The author proved that neuro fuzzy systems could be very
effective in long-term prediction of this type of traffic. Izquierdo et al. [45] in 2007, proposed
neuro fuzzy approaches for dynamic classification of the road vehicles to avoid road vehicles
collision. In the same year, Partoucheet al. [46] presented a new approach to autonomously adapt
the speed of a vehicle by learning from a human driver and using anticipation. Later, Aseri and
Bagai [47] proposed ANFIS to solve the problem of rate control for Available Bit Rate (ABR)
service class in Asynchronous Transfer Mode (ATM) networks. Sindal et al. [48]in 2009
developed neuro fuzzy call admission control algorithm to reduce voice data traffic in CDMA
cellular network by transmitting signals as lower power level. In 2011, Abbas et al. [49]
proposed neuro fuzzy system for best route selection to avoid traffic congestion. They used
neuro fuzzy logic and ant colony system (ACS) algorithm for routing. The researcher’s
motivation behind the research work was to avoid road traffic, road accident, network
congestion, video traffic and vehicles congestion. The methodology, domain and article type of
traffic control categorized in Table 4.4

Table 4.4. Traffic control

6. NFS in image processing and feature extraction

Neuro fuzzy system has a wide range of applications in the imaging analysis. Imaging analysis is
the process of extraction data or information from images by means of image processing
techniques. Computers are indispensable for the analysis of large amounts of data which contains
the fields of computer or machine vision and medical images for tasks that require complex
computation for the extraction of quantitative information and makes use of pattern recognition,
digital geometry, and signal processing. Some applications related to image processing and
feature extraction include emotion recognition, image analysis, noisy image processing, face
recognition and image compression. A brief description of works done by various authors in the
last decade is presented below. In 2005, Ioannou et al. [50] created rule based neuro fuzzy
system to describe an emotion recognition system. This system combined psychological findings
about emotion representation with analysis and evaluation of facial expressions. The authors
classified facial expressions using a continuous 2D emotion space. Later in 2007, Thanh et al.
[51] proposed a generalized fuzzy inference system (GFIS) in noise image processing which was
a multi-layer neuro-fuzzy structure that combined both Mamdani model and Takagi–Sugeno (T–
S)fuzzy model to form a hybrid fuzzy system.

Table 4.5 Image processing and feature extraction

GIFS acquired the interpretability property of the Mamdanimodel and stability criteria of
Takagi–Sugeno model. In the same year Liu and Sung [52] presented a scheme based on feature
min-ing and neuro-fuzzy inference system for detecting LSB matching steganography in gray
scale images. Support vector machine recusive feature elimination (SVM-RFE) was proposed to
obtain better detection accuracy. Chakrapani and Soundararajan [53] proposed ANFIS for fractal
image compression (FIC) in spatial domain in 2009.Authors used ANFIS based network for
classifying the domain cells of a gray level image based on its statistical characteristics to per-
form fractal image compression. This technique incorporated the concept of NN in creating the
rules and produced a fuzzy model based on Takagi–Sugeno approach. In that year, Makhsoos et
al. [54]did their research work on face recognition using multi layer perceptron (MLP) network
by combining fuzzy logic, neural network and mixture of experts. They presented the importance
of combining these two technologies (neural networks and fuzzy logic) in face recognition.
Marichal et al. [55] in their paper presented an approach based on a simulated environment and
the application of a Neuro-Fuzzy technique in order to extract features of the environment by
ultrasonic sensors while a vehicle is moving. In the year 2010, Gomathi et al. [56] proposed
neuro fuzzy approach for facial expression recognition system to recognize the human facial
expressions like happy, fear, sad, angry, disgust and surprise using local binary pattern (LBP)
histogram. This model reported 95.29% of classification accuracy. Montazer et al. [57] proposed
a neuro-fuzzy inference engine to recognize the Farsi numeral characters. Lin and Chen [177]
developed an application using fuzzy models for the monitoring of ecologically sensitive
ecosystems in a dynamic semi-arid landscape from satellite imagery. Sancheza et al. [132]
introduced a novel neuro-fuzzy system FasArt(Fuzzy Adaptive System ART-based) for both
clustering and classification purpose which was tested on a dataset produced by UNIPEN
project. Motivation of the researchers behind this research work is to reduce the noise in image,
face recognition, image compression, facial expression recognition, character recognition,
analysis in image etc. The methodology, domain and article type of image processing and feature
extraction are categorized in Table 4.5.

7. NFS in forecasting and prediction

Forecasting and prediction is the process to predict future events and conditions and should be
key decision-making elements for management in service organizations. The term ‘forecasting’
is sometimes reserved for estimates of values at certain specific future times, while the term
prediction is used for more general estimates of values over a long period of times. During the
last decade a huge number of researchers have contributed their innovations in this category and
a number of research works have been conducted in recent years. In 2005, Nayak et al. [60]
proposed an adaptive neuro fuzzy inference system for river flow forecasting. Authors
demonstrated the applicability of neuro fuzzy systems in developing effective nonlinear models
of the rainfall-runoff process for short-term flood forecasting. Ang and Quek [61] presented a
novel rough-set based neuro-fuzzy approach for forecasting stock trading difference. Linet al.
[62] in 2008 proposed ANFIS based approach to predict sea level using tide generating forces
and sea surface temperature. The proposed model named tide-generating forces considering sea
surface temperature and fuzzy neuro-network system (TGFT-FN).This paper presented an
adaptive neuro fuzzy inference system for predicting sea level considering tide-generating forces
and oceanic thermal expansion. Authors assumed that sea level is dependent on sea surface
temperature. Chen et al. [178] in 2009, constructed a mathematical model for ocean environment
to consider the influence of the external waves on oceanic structures. They employed Takagi–
Sugeno fuzzy model in the approximation of the oceanic Structure and performed stability
analysis using the Lyapunov direct method. Next year, Chen et al. [183] proposed a Takagi–
Sugeno fuzzy model based controller, which can attenuate the influence of the external wave
force. The controller was evaluated in terms of stability analysis and the linear matrix inequality
(LMI) conditions. Later in 2011, Patil et al. [63] presented ANFIS to predict ocean wave
transmission. Authors implemented ANFIS using neural network and back propagation
algorithm. Chen [136] presented a Takagi–Sugeno fuzzy model for the modeling and stability
analysis of oceanic structures. Author discussed vibration of the oceanic structure analytically
based on fuzzy logic theory, fuzzy Lyapunov theory and a mathematical framework. Tsenget al.
[140] described a critical assessment of risk control decision model based on interpretive
structural modeling (ISM) and

Table 4.6 Forecasting and predictions

decision making method. ISM is an interactive learning process in which a set of different and
directly related elements are structured to form a comprehensive systemic model. Lin and Chen
[170] extended the Takagi–Sugeno fuzzy model representation and Lyapunov method to analyze
the stability of community and ecosystem hierarchies based on a framework of ecological
hierarchy. Recently ,Sudheer and Mathur [64] integrated ANFIS with particle swarm
optimization algorithm to achieve steady uniform flow optimal solution to a ground water flow
and contaminant transport problem. Galavi et al. [65] in 2012 proposed ANFIS based method
forwater resource forecasting. They used hybrid learning algorithm for this purpose. Lin et al.
[144] developed a fuzzy-rule-based risk assessment model of debris flows to verify the accuracy
of the risk assessment in order to help related organizations reduce losses caused by debris flows
in Taiwan. Authors adopted a fuzzy-rule-based decision system to determine and verify the risk
of debris flows. Liu et al. [182] summarized the criteria of significance evaluation from the past
review results and utilized fuzzy rule-based system to incorporate these criteria into scientific
facts. They also employed data mining technique to construct an environmental impact statement
for reviewing results. Besides these, a set of articles [155,156,162,169,179,185] can be included
in this category, which mainly concerns about bridge health monitoring, structural damage,
seismic hazard, risk analysis etc. An adaptive neuro fuzzy inference system has established itself
as one of the popular modeling technique in this field. Their motivation behind the research work
was to predict and forecast in different categories-such as, water resource forecasting, sea level
prediction, oceanic structures, risk control etc. The methodology, domain and article type of
fore-casting and prediction are categorized in Table 4.6.

11. NFS in social sciences

Knowledge required for social regulation can be of two types: explicit and tacit [130]. Explicit
knowledge relates to the community culture indicating how things work in the community
basedon social policies and procedures. Tacit knowledge is ethics and norms of the community.
The former could be codified, stored and transferable in order to support decision making, while
the latter being based on personal knowledge, experience and judgments is difficult to codify and
store. As per our knowledge no such contribution of NFS is found in social sciences since last
decade. Lazareyica et al. [129] in 2004 proposed a neuro-fuzzy modeling to support knowledge
management in social regulation. They investigated neuro-fuzzy support of knowledge
management in social regulation. In their model, the explicit knowledge was based on social
policies and procedures to reduce smoking among youngsters.