Sustainable Computing: Informatics and Systems 20 (2018) 192–202

Contents lists available at ScienceDirect

Sustainable Computing: Informatics and Systems

journal homepage: www.elsevier.com/locate/suscom

Intelligent computing system based on pattern recognition and data

mining algorithms
Junlin Zhang a,∗ , Samuel Oluwarotimi Williams b , Haoxiang Wang c,d
a
Xi’an University of Architecture and Technology, Xi’an, China
b
Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Xili Nanshan, Shenzhen, 518055, China
c
Cornell University, NY, USA
d
R&D Center of GoPerception Laboratory, NY, USA

a r t i c l e i n f o a b s t r a c t

Article history: The integration of intelligent system mainly includes the application of intelligent technology, such as
Received 28 July 2017 artificial intelligence and computational intelligence method, which is used in different levels of the
Received in revised form 21 October 2017 system. This paper introduces the application and technology of several intelligent system integrations,
Accepted 22 October 2017
the advantages and disadvantages of learning theory and expert system. Neural network is applied in
Available online 11 November 2017
intelligent systems and we use scope reviewed several new development of intelligent technology, plus
this paper describes the development direction of the intelligent system. This paper introduces the basic
Keywords:
concepts of data mining, including data mining technology, artificial intelligence, machine learning, sta-
Pattern recognition
Data mining
tistical analysis, fuzzy logic, pattern recognition and artificial neural networks and other technologies. We
Intelligent systems analyze the structure of the general algorithm of data mining, and classify the data mining technology in
Technologies details, including more than 10 techniques of decision tree technology, neural network technology, rough
Algorithms set and fuzzy set. Finally, the research directions of data mining in artificial intelligence, e-commerce
applications and mobile communication computing are discussed.
© 2017 Elsevier Inc. All rights reserved.

1. Introduction must be a comprehensive fringe science which combines calcula-

In recent years, with the continuous development of intelligent
calculation method, the model of intelligent computing meth-
ods in dynamic optimization application increased significantly,
including ant colony algorithm, genetic algorithm, particle swarm
algorithm, differential evolution algorithm, cultural algorithm and
so on. The above calculation method does not need to rely on gradi-
ent information, which has a wide range of use, and the calculation
method does not need to set the initial point, so it is easy for us to
operate. These intelligent computing methods are often based on
population probability search, in general, they will not fall into local
extremum. Therefore, the intelligent computing method can solve
the limitations of the traditional calculation method through the
system dynamic optimization method, and improve the accuracy
and efficiency of solving the dynamic optimization problem [1,2].
Artificial intelligence is a branch of computer science. It is
applied to many fields, such as expert system, pattern recognition,
neural network and so on. From the point of view of its functions, it
∗ Corresponding author.
E-mail address: zhangjunlinxian@126.com (J. Zhang).
tion, control theory, information technology, linguistics and other
disciplines. In addition to numerical calculation and data process-
ing, another important application of computer is the development
and application of artificial intelligence. How to use computers
to simulate part of human intelligence activities is the research
direction of artificial intelligence systems. The cognitive system of
artificial intelligence is based on logic, and its deductive reasoning
is performed by symbolic processing [3,4]. The intelligent system is
relatively fragile and lacks stability, which is also sort of the key and
difficult point for the future development of intelligence (Fig. 1).
The theory of intelligent computing is mainly based on con-
nectionism, and it intersects with the mathematical methods such
as fuzzy mathematics and iterated function system, which has
formed many developing directions. Artificial neural networks,
genetic algorithms, evolutionary computation, artificial life, eco-
logical computing, immune information processing, multi-agent
systems, etc. can all be included in intelligent computing. In
the study of human intelligence behavior, most human activities
involve more, and the solution of large complex problems needs
collaborating by many professionals or organizations. “Collabora-
tion” is one of the main forms of human intelligence behavior, and
distributed artificial intelligence is about to meet this demand. With

https://doi.org/10.1016/j.suscom.2017.10.010
2210-5379/© 2017 Elsevier Inc. All rights reserved.
 J. Zhang et al. / Sustainable Computing: Informatics and Systems 20 (2018) 192–202
Fig. 1. Intelligent computing system.
Fig. 2. Heterogeneous processor: Netronome x86 + NFP.
the development of computer network and programming, dis-
tributed artificial intelligence has become a new research hotspot
in artificial intelligence field. The main research purpose in logic or
intelligent distributed dynamic author is figuring how to coordinate
their behavior, and how to coordinate their knowledge, skills and
planning, in order to solve the single objective or multi-objective
problem, and to provide an effective way to support collaborative
work of large and complex intelligent systems or computer design.
The nature of the distributed system determines is complex, non-
linear, and it achieves a higher order state through coordination
among subsystems [5,6].
Intelligent computing is a comprehensive technical system
which integrates many advanced technologies and latest trends
in the field of IT infrastructure, and it has the characteristics of
automation, integration and security. It includes large-scale data, as
well as emerging services delivery models such as cloud computing
(Fig. 2).
Along with the development of the connection technology,
recently the industry is an important trend from the original com-
puter to the computing ability of transformation, which is now no
longer adhere to a device, requiring more emphasis on the exis-
tence of computing power [7]. This trend has affected our lives in
every way, and the trend has created a variety of new products,
such as digital signage, kiosks, and other infrastructure connected
devices. The most critical thing is the different equipment, which
has its own computing power, on the other hand, they are con-
nected together, providing us with new opportunities and new
needs. Embedded system has been widely used in the entire elec-
tronic industry, involving control systems, calculation and other
processing applications.
193
2. Intelligent computing system
Based on the comprehensiveness of artificial intelligence sys-
tem, its development is bound to be the integration of many
intelligent technologies. Intelligent information processing system
of human-machine systems can help us make scientific decisions
in complex computing problems, and the application of expert sys-
tem technology in this area has been very skilled in various fields
resulting some remarkable achievements. But the most important
problem of the development of expert system is knowledge acqui-
sition and processing. The knowledge of each subject and the strict
distinction between them make it difficult to deal with the intel-
ligent system. Therefore, integration of intelligent technology and
expert system, such as fuzzy logic and neural networks, has become
an inevitable trend [5,8].
On the basis of the theories of statistics, set theory, information
theory and artificial intelligence, people put forward a variety of
data mining techniques and methods based on intelligent comput-
ing, and formed their own characteristics and general application
fields. In general there are several basic techniques: the imitation
biotechnology, statistical methods, collective methods, decision
trees, fuzzy systems, cloud theory, self-organizing mining tech-
nology. These technologies have the following inner features. (1)
The evolution strategy is optimized to one of the functions and
then carry on the evolution operation to obtain the optimal solu-
tion, and it optimizes the function to take the maximum value or
the minimum value which is not limited. Firstly, determined the
evolution object, while namely treats optimized the variable. How
two parts calculate the evolution function and the general ground
may use the ES algorithm in the objective function cluster ques-
tion, however we usually only can obtain the approximate optimal
solution and the overall situation optimal solution. (2) The self-
organizing feature mapping network method is an unsupervised
clustering method, which gathers data through repeated learning
to compete with the current object through several units. The unit
which is close to the current object becomes an active or winning
unit. To get closer to the input object, adjusting the weight of the
winning cell and its nearest neighbor. (3) Group intelligence pro-
vides a basis for finding solutions to complex distributed problems
without centralized control or providing a global model. The main
idea is to use the group intelligence algorithm to cluster, and to
reduce the knowledge of the data vector group randomly into a
two-dimensional plane, and then to produce some virtual ants in a
given plane. (4) Realistic data is usually complex or mixed, struc-
tured and unstructured, and incomplete, which is characterized,
such as these complex data sets. The simple distance measure is
not enough to measure the similarity or distinguish important fea-
tures. Therefore, we need to find new measures and design based
on the new measure of clustering analysis methods in order to deal
with complex data sets.
2.1. Electric power system
The development trend of power system automation is monitor-
ing into to the open-loop closed-loop control, where high voltage is
extended to low voltage. Among them, the main application of arti-
ficial intelligence technology is the diagnosis of power equipment
fault [9,10].
At present, the main technologies used in power system fault
diagnosis are ANN/ES/GA/Petri/FST and so on. The core problem
of fault diagnosis is information extraction, and the uncertainty of
information increases the difficulty of information extraction. To
solve this problem, there are expert system based fault diagnosis,
neural network based fault diagnosis and fault diagnosis based on
optimization technology.
194 J. Zhang et al. / Sustainable Computing: Informatics and Systems 20 (2018) 192–202
Expert system can provide power emergency treatment, system
restoration, system planning, fault isolation, static and dynamic
security analysis, etc. The results are satisfactory for the complete
and accurate signals, but the limitation of fault tolerance is still
far from the required results, especially when the information is
incomplete, thus the diagnosis of power systems is prone to errors.
Neural network has nonlinear characteristics, and its parallel pro-
cessing ability, robustness and self-organizing learning ability have
been pursued by researchers in various fields for two years. Neural
network algorithm is more suitable for small and medium power
system failures [11].
Based on optimized fault diagnosis, genetic algorithm is used to
optimize the system diagnosis. Genetic algorithm needs to estab-
lish the model of fault point, and the prediction ability of the
warning information can predict the location of the fault point. We
can give an fitness function that makes the diagnosis an integer
programming problem.
2.2. Search engines
Searching for information is not our goal, and only when we find
information useful to us can we say that the search is successful. The
development of intelligent search engine is to solve this problem. It
has considerable knowledge processing ability and understanding
ability. When keyword based search does not satisfy search require-
ments; the requirement of intelligent search engine is raised to the
knowledge based level.
The intelligent search engine can work across platforms, and
its ability of processing a variety of mixed documents allows us
to accurately retrieve the latest information during the uninter-
rupted information replacement process. Intelligent search engine
is initiative, it can observe the behavior of the user actively, and
understand the focal point of the user and from the user’s point
of view. Through continuous improvement of learning, improving
search accuracy. Personalized search is an important breakthrough
of intelligent search engine, and it is an important way to improve
search accuracy [12,13]. The search engine for users to search the
effective classification information can make the search engine to
meet each customer needs, and allow customers to customize web
page, then select the items of interest or frequent landing site.
The intelligent search engine should have the knowledge base
and the information base. The implementation of personalized
retrieval and retrieval learning requires the use of information
retrieval, computer networks, natural language processing, dis-
tributed artificial intelligence, automatic theorem proving and
other theoretical techniques.
The intelligent agent technology can integrate the client’s
special environment and complete the search according to the
user’s interest. Reduce workload during user search. Through
machine learning, users can independently and independently find
databases of interest to users. These databases can be divided into
different regions and improved retrieval performance. You can also
monitor information sources in real time so as to reduce retrieval
time. Net to net technology, that is, the application of neural net-
work technology, which must establish a stable data model for Web
to obtain information. Web mining technology is a pattern of dis-
covering interest points from information sources and activities.
It includes content, structure and access information mining. The
application of mining technology can improve the accuracy and
generalization rate of query.
2.3. Video monitoring system
Video surveillance technology is applied in enterprise produc-
tion, life and so on. It can be used in real-time alarm system
and scene detection. Real time alarm system is divided into user-
defined alarm and abnormal behavior automatic alarm. It has broad
application prospects for museums, traffic, road behavior, station
monitoring and other places. Scene detection is divided into the
number of personnel detection, traffic detection and production
process detection, and is widely used in industry [14,15].
To realize intelligent video surveillance, we should adopt several
new technologies including moving object detection technology,
motion tracking technology, automatic video retrieval technology,
pattern analysis technology.
2.4. Knowledge processing methods contained in collaborative
intelligent computing model
Intelligent computing, in essence, is sort of connection and
information interaction under the environment of new processing
platform, involving sensing, data acquisition, data analysis and pro-
cessing, a large system simulation technology and communication
connection.
The combination of intelligent computing and mobile devices
is the general trend in the future. At the latest Taipei computer
show, more and more mobile PC are following the footsteps of
mobile phones to realize many functions. In the calculation of intel-
ligence, transmission speed is the responsibility of the network, and
in the embedded system, energy saving and high performance of
the balanced development increase the performance of portable
processing products, which will provide more processing power
and the execution of software to obtain information, and let the
intelligent and energy saving more thorough [16,17].
One of the core concepts of intelligent computing is that the ter-
minal collects various information and collects it into a data center.
Then, the intelligent algorithm is used for data analysis and mining
to achieve the goal of intelligence. The typical application of embed-
ded and network is a vending machine, by sending the information
collected by the different devices to the client, not only can the
data analysis submitted to the supplier for targeted delivery but
also provide different crowd users demand, so as to give targeted
procurement proposals.
3. Pattern recognition
When people see something or phenomenon, people will first
collect all the information of the object or phenomenon, then the
behavior and mind has some related information to compare to, if
you find a same or similar matching, people can recognize the object
or phenomenon. Therefore, the relevant information of an object or
phenomenon, such as spatial information, temporal information,
etc., constitutes the model of the object or phenomenon. Broadly,
things that can be observed in time and space can be called patterns
if they can be distinguished from whether they are the same or
similar [16,17].
Human beings have strong pattern recognition ability. It
through the visual information to identify text, pictures and the
surrounding environment, through the auditory information iden-
tification and understanding of language to finalize the PR process.
Pattern recognition is a basic cognitive ability or intelligence of
human being. It is an important component of human intelligence
and plays an important role in all kinds of human activities. In real
life, almost everyone can easily accomplish the process of pattern
recognition. But I’m afraid it’s not that easy if you want the machine
to do the same thing. From the point of view of artificial intelligence,
this paper analyzes the concept of pattern recognition and how to
recognize patterns by machine.
Pattern recognition is a mathematical model that studies human
recognition, which uses computer technology to allow computers
to model the behavior of human recognition. In other words, pat-
 J. Zhang et al. / Sustainable Computing: Informatics and Systems 20 (2018) 192–202
tern recognition is the study of how to let the machine look at the
environment, and to learn to identify patterns of interest from the
background, and to make accurate judgments about the class of the
pattern. Recognition behavior can be divided into two broad cat-
egories: identifying specific things and identifying abstract things.
The identification of specific things involves the identification of
spatio-temporal information. The identification of abstract matters
involves the identification of a problem, solution, or argument. In
other words, the recognition of abstract things is the identifica-
tion of phenomena that does not exist in the form of matter, which
belongs to the category of conceptual recognition research. Pattern
recognition refers to the identification of specific things, such as
speech waveform, seismic wave, ECG, EEG, pictures, text, symbols.
Three-dimensional objects can be measured by physical, chemical,
biological and other specific patterns.
3.1. Pattern recognition system
A complete pattern recognition system consists of data acqui-
sition, data processing, feature extraction and selection, and
classification decision making [18,19].
In the design of pattern recognition system, we need to pay
attention to model class definitions, applications, pattern rep-
resentation, feature extraction and selection, clustering analysis,
classifier design and learning, training and testing sample selection,
performance evaluation etc. For different application purposes, the
content of each part of the pattern recognition system can vary con-
siderably, especially in data processing and pattern classification. In
order to improve the reliability of the identification results, we need
to add the knowledge base to correct possible errors, or by introduc-
ing constraints which greatly reduce the pattern recognized in the
model library of the search space, in order to reduce the matching
calculation. In some specific applications, such as machine vision, in
addition to identifying what the object is, the position and posture
of the object must be determined to guide the robot’s work.
3.1.1. Data acquisition
Data acquisition refers to the use of a variety of sensors to con-
vert the various information of the object into a set of values or
symbols that the computer can accept. We call this kind of numer-
ical or symbolic (string) space as the model space. The key to this
step is the selection of sensors. In order to extract valid information
from these numbers or symbols, data processing must be carried
out, including digital filtering and feature extraction.
3.1.2. Data processing
Data processing is to eliminate the noise in the input data or
information, and eliminate the irrelevant signals, leaving only the
features and properties of the subjects and the identification meth-
ods are closely related (such as the representation of the object’s
shape, perimeter, area etc.). For example, in fingerprint recogni-
tion, fingerprint scanning equipment for each output with image
contrast, and brightness or background are different, sometimes
they may be deformed. Therefore, it is necessary to adopt appro-
priate filtering algorithms, such as directional filtering based on
block diagram, two valued filtering, etc., in order to filter out these
unnecessary parts in the fingerprint image [20].
3.1.3. Feature extraction
Feature extraction refers to derive useful information from data
filtering, to find out the most effective features from many features,
in order to reduce the processing difficulty of human features easy
access to machines, which is difficult to obtain. Feature selection
and extraction are key problem in pattern recognition. In general,
the more types of candidate features, the better results should be
obtained. However, it may lead to the curse of dimensionality,
195
that is, the feature dimension is too high to be solved by com-
puter. Therefore, the key of data processing stage is the selection
of filtering algorithm and feature extraction method. For different
applications, the filtering algorithm and feature extraction method,
as well as the extracted features will also be different [21].
3.1.4. Classification decision or model matching
Based on the pattern feature space generated by data process-
ing, we can carry out the last part of pattern recognition: pattern
classification or model matching. The final output at this stage may
be the type of object to which it belongs, or it may be the model
number that is most similar to the object in the model database. A
pattern, classification, or description, usually based on a collection
of patterns that have been classified or described and this model
is called a training set, and the resulting learning strategy is called
supervised learning. Learning can also be unsupervised learning. In
this sense, the system does not need to provide a priori knowledge
of the pattern class, but it is based on the statistical laws of the
patterns or the similarity of the patterns.
Member table: template matching. Based on this idea, pre stored
process belongs to the same class, and the unknown mode and sys-
tem input is compared with the same or similar pattern, namely
the unknown pattern type.
General characteristics: general patterns are stored in a classifi-
cation system, when there is an unknown pattern into the system,
the system will be compared with the general characteristics of the
general characteristics and existing classes in the system, and put
it into the class with similar characteristics [22,23].
Clustering: if the target vectors are far apart from each other in
geometry, it is easy to determine the class of unknown patterns.
However, if the target vectors are closer or even overlap, people
need to adopt relatively complex algorithms to determine the class
of unknown patterns. Minimum distance classification is a simple
algorithm based on the concept of clustering. By calculating the
unknown pattern, to decide which one belongs to known model
and the unknown model recently, and finally the unknown patterns
are known as model classes. The algorithm is very effective for clas-
sifying the target vectors at different distances from the geometric
positions.
Neuron: bionics refers to the application of biological knowl-
edge to electronic machines. The neural system approach
introduces the biological knowledge to the machine for pattern
recognition, thus introducing artificial neural networks. A neural
network is an information processing system consisting of a large
number of simple data processing units, which works together to
achieve large-scale parallel distributed processing. The design and
function of a neuronal network is designed to mimic the biological
brain and nervous system functions [24,25]. Neural networks have
the advantages of adaptive learning, self-organization and fault tol-
erance. Because of these prominent features of neural networks,
neural networks can be used for pattern recognition. Some of the
best neural network models are backward propagation networks,
higher order networks, delay and periodic networks.
In general, people use forward propagation networks for pattern
recognition. Forward propagation is feedback that does not return
to the input. Similar to what humans have learned from mistakes,
neural networks can learn from their mistakes by feeding back
information to the input. Through feedback, the input pattern can
be reconstructed to avoid errors and improve the performance of
the neural network. Obviously, the construction of such a network
of neurons is very complex. The back-propagation algorithm is used
in this kind of neural networks. One of the main problems of the
backward propagation algorithm is the local minimum problem. In
addition, neural networks have some problems in the aspects of
learning speed, structure selection, feature representation, modu-
larity and scaling. Although there are such problems and difficulties
196 J. Zhang et al. / Sustainable Computing: Informatics and Systems 20 (2018) 192–202
Table 1
Four basic types of fuzzy pattern recognition.
Standard Sample to be Methods
mode identified
R1 , ...Rc X Maximum membership principle
R1 , ...Rc X Proximity principle
∼
R1 , ...Rc X Proximity principle/Synthetic fuzzy set
R1 , ...Rc X Synthetic fuzzy set/Synthetic approach degree
∼
in neural networks, there is still great potential for the development
of such networks [26].
3.2. Pattern recognition method
Definition 1. Assume we have n fuzzy subsets in discourse
domain, denoted as follows:
An = {A1 , ..., An }
Ui {i = 1, 2, ..., n}
We call fuzzy vectors as fuzzy vector sets, denoted as:

n
Fn ≡ {A} = F(Ui )
i=1
Definition 2. A set of fuzzy vector sets is selected randomly,
denoted as A ∈ Fn . We build a synthetic fuzzy set, denoted as
A≡ < A > ≡ < A1 , A2 , ..., An >. Membership function is calculated as:
A(u) = Mn (A1 (u1 ), A2 (u2 ), ..., An (un ))
Whereu = (u1 ,u2 , ...,un ) ∈ U∗.
Axiom of nearness: A and B are two fuzzy subsets, mapping
satisfies the following equations:
 : F(U) × F(U) → [0, 1], (A, B) → (A, B)
(A, A) = 1
(A, B) = (B, A)
A ⊆ B ⊆ C ⇒ (A, C) ≤ (A, B) ∧ (B, C)
3.2.1. Four basic models of fuzzy pattern recognition
The fuzzy pattern recognition problem can be divided into the
following four basic models from the perspective of the identified
object (Table 1).
3.2.2. Artificial neural network based pattern recognition
In 1950, F.Rosenblatt proposed a simulation of the human brain
based on a simplified mathematical model, and perception, recog-
nition, preliminary implementation training and recognition ability
are included in the system. In 1980, J.Hopfield revealed Lenovo stor-
age and computation capacity of artificial neural network, which
is a new approach for pattern recognition technology, so as to
form the artificial neural network pattern recognition method.
Neural pattern recognition takes advantage of neural computing
patterns that arise in neural networks. Most neural networks have
some training rules, such as adjusting connection weights based
on existing patterns. In other words, the neural network learns the
examples directly and obtains its structural features for generaliza-
tion [18].
Artificial neural networks can outperform traditional computer-
based pattern recognition systems. People can recognize patterns
by using computers or neural networks. The computer uses tradi-
tional mathematical algorithms to detect whether a given pattern
matches the existing pattern. This is a simple and easy understood
method. However, the method can only be judged by “yes” or “no”
which does not allow the model to have noise. On the other hand,
neural networks allow patterns to be noisy, if properly trained,
neural networks can respond correctly to categories of unknown
patterns. For example, BP neural network directly learns observa-
tion data which is very simple and effective, so it has been widely
used, but it is a heuristic technology, without a solid theoretical
basis for specifying engineering practice.
Fuzzy pattern recognition and neural network pattern recogni-
tion are newly developed pattern recognition methods, and they
are important components of information science and artificial
intelligence. In the past few decades, people are interested in fuzzy
mathematics, thus artificial intelligence and rule-based expert sys-
tems has soared. Pattern recognition plays an important role in
these research fields [28].
The following table describes some of these pattern recognition
methods. In fact, the pattern recognition method is not completely
independent of each other. In many emerging applications, where
(1) there is no optimal approach, several different pattern recognition
methods must be used simultaneously. Attempts have been made
(2) to design a pattern recognition system incorporating a number of
identification methods (Table 2).
With the rapid development of computer hardware and soft-
ware technology, pattern recognition has gained more and more
(3) attention, the pattern recognition technology is becoming more
and more perfect, and has been successfully applied in many fields,
such as data mining, document classification, financial forecast-
ing, multimedia database organization and retrieval, biological
(for example according to the physical characteristics, such as
human face and fingerprint identifying people), medical (med-
(4) ical image analysis), energy, geology, meteorological (weather
(5) forecast), chemical industry, metallurgy, aviation (satellite aerial
photograph interpretation), and the field of industrial product test-
ing etc. The field of pattern recognition is the fastest development
in recent years which should belong to the computer visual and
(6) auditory fields, such as handwriting recognition, biometric identi-
fication (including: fingerprint recognition, iris recognition, retina
(7) recognition, palmprint recognition, face recognition, palm vein
recognition, distribution).
(8)
Pattern recognition is a fast developing discipline, so it is diffi-
(9) cult to make a comprehensive and detailed summary of the latest
research progress in this field. With development of pattern recog-
nition, there have been many effective pattern recognition methods
to solve different problems, but they have not yet developed into
a unified and effective pattern recognition theory for all problems.
The purpose of pattern recognition is to develop general data anal-
ysis techniques that do not rely on application domains, so that
machines can be analyzed and solved like human beings. This is a
difficult goal, the current work is to figure out a combination of the
specific problems and propose new methods of pattern recognition.
3.2.3. Template matching pattern recognition
The principle of template matching is to select the known object
as a template, and to compare the selected area with the image to
identify the target. The computation of template matching is very
large, and the storage of the corresponding data is also very large.
Moreover, as the image template increases, the amount of com-
putation and storage increases with the geometric number. If the
image and template are large enough, the computer will not be able
to deal with it, and then it will lose the meaning of image recog-
nition. Another disadvantage of template matching is that it can
achieve the optimal solution in theory because of many matching
points, but it is very difficult to achieve in practice.
 J. Zhang et al. / Sustainable Computing: Informatics and Systems 20 (2018) 192–202 197

Table 2
Common pattern recognition methods.

Methods Expression pattern Identification function Criterion

Template matching Sample, pixel, curve Correlation, distance measurement Classification error
Statistical pattern recognition Feature Generic discriminant function Classification error
Structure pattern recognition Element Rules, grammar Acceptance error
Fuzzy pattern recognition Feature Membership function Membership degree
Neural network pattern recognition Sample, pixel, feature Nonlinear signal processing function Mean square error

Template matching is mainly used in the detection of object coordinate.The kinetic equations used for image recognition are
position in images, tracking of moving objects, registration of described as follows:
images between different spectrums or different photography time.   
q= k vk (vk q) − Bkk (vk q)2 (13)
k k =
3.2.4. Pattern recognition based on support vector machines
Support vector machine is a new kind of classification tech-
nology proposed in 1963, its basic idea is: first, based on the
sample space, constructing an optimal hyper-plane to maximize
the distance between the hyperplane and different samples, so
as to achieve the maximum generalization ability. Support vector
machines (SVM) are simple in structure, and have global optimal-
ity and better generalization ability. They have been widely studied
since they were introduced.
Support vector machines (SVM) is an effective tool for solving
pattern recognition and function estimation problems. SVM in dig-
ital image processing applications is to find the difference between
the image pixels from the pixel, feature and surrounding environ-
ment itself (adjacent pixels).
3.3. Synergetic pattern recognition
The basic idea of synergetic is that a high-dimensional nonlin-
ear problem is reduced to a nonlinear equation with the same set of
dimensions. The order parameter equation controls the dynamics
of the system near the critical point. By solving the order param-
eter equation, we can obtain temporal, spatial or spatiotemporal
structure.
Assumeastatevectordescribessystemstatus, denotedasq = (q1 , ...,qn )(10)
In all cases considered by mathematical theory of synergetic, the
state vector of the time derivative follows the following equation:
q(x, t) = N[q(x, t), ∇ ˛, x, t] + F(t) (11) uses the results of this analysis to define the parameters of the min-
The dependence of the stable modulus is proved by the dom-
ination principle of synergetic. In the process of system motion,
the stable modes are gradually weakened, and some unstable
modes are enhanced, which become the main structural factors
of the system. The magnitude of the unstable modes is called the
order parameter. In essence, the final state of the system will be
determined by the unstable modulus with the largest initial order
parameter. Therefore, the order parameter of unstable modes can
be used only. The modeling of synergetic is mainly to establish
stochastic differential equations. The stability principle is used to
eliminate the stable modulus so as to obtain the closed equation of
order parameter. Thus the high-dimensional problem is reduced to
a low dimensional problem.
The process of pattern recognition corresponds to a kinetic pro- accuracy of prediction. The commonly used model representation
cess. Assume a virtual particle describing a pattern moves on a
potential terrain, and when a particle enters a certain attractive
valley, a corresponding pattern is identified. This process can be
described as:
q(0) → q(t) → vk (12)
In image pattern recognition, the image matrix is transformed
into one dimension vector. We do not consider the spatial infor-
mation, so the state vector is only related to the temporal
/ k / k
k=
(vk q)vk − C(q+ q)q + F(t) (14)
Compared with the traditional pattern recognition system,
cooperative pattern recognition system reduces the process of
feature extraction and selection. The recognition is based on the
method of collaborative prototype model, so the recognition ability
of collaborative method mainly depends on the prototype model,
and collaborative algorithm can only base on further optimization
of synergetic neural network in other aspects. Therefore, collab-
orative pattern recognition method is more suitable for solving
license plate recognition, fingerprint identification, face recogni-
tion, industrial parts recognition and so on.
In order to further improve the ability of image recognition and
classification system, we can adopt collaborative method and tra-
ditional recognition method based on pattern recognition, feature
extraction. Synergetic neural network for recognition learning is
also a direction of collaborative pattern recognition method. For
the problem of invariant recognition, we can extract the invariants
of the image in the spatial variation, and recognize the invariants
by cooperative pattern recognition, so as to achieve the goal of
invariant recognition of the target pattern (Fig. 3).
4. Data mining algorithms
Data mining algorithms are mechanisms for creating data min-
ing models. In order to create a model, the algorithm first analyzes a
set of data and looks for specific patterns and trends. The algorithm
ing model. These parameters are applied to the entire data set for
extracting feasible patterns and detailed statistical information.
Data mining is closely related to knowledge discovery. Knowl-
edge discovery refers to the whole process of discovering useful
knowledge from databases. It includes data selection, preprocess-
ing, data transformation, data mining, schema interpretation and
knowledge evaluation. Data mining is a key step in the process of
knowledge discovery. Data mining is the extraction of useful infor-
mation patterns from large amounts of random data. The purpose
of data mining is to improve the market decision-making ability.
Model representation used to describe the model as a language.
If the language is descriptive, it helps to find accurate mathematical
models. However, the descriptive language which is too powerful
may lead to the over generalization of the model and reduce the
methods include decision tree, nonlinear regression, case-based
reasoning, Bayesian network and inductive programming.
Model evaluation criteria is to predict the class model, some
test data sets can be used to evaluate its accuracy. The model of
description class can be evaluated in terms of accuracy, novelty,
practicability and understandability.
Detection methods are divided into parameter discovery and
model discovery. After the model representation and model eval-
uation criteria have been determined, data mining has become an
198 J. Zhang et al. / Sustainable Computing: Informatics and Systems 20 (2018) 192–202
Fig. 3. Structure representation of chromosome of parameter optimizationalgorithm of synergetic neural network.
Fig. 4. Data Mining Algorithm.
optimization task, namely, finding the parameters or models most
suitable for the evaluation criteria from the description of the data.
In particular, parameter discovery is to find the parameters that
are most suitable for the model evaluation criteria after determin-
ing the data set and model representation. Model discovery is a
cyclic heuristic process, and we need to constantly change model
representations and finally determine the appropriate number of
models (Fig. 4).
In general, there is no universally applicable algorithm. An algo-
rithm is very effective in one area, but it may not be appropriate
in another. Therefore, in practical applications, we must select
effective data mining algorithms for specific fields. In other words,
the actual development often turns into formalization of domain
knowledge and discovery tasks, rather than optimizing the selected
data mining algorithms in detail.
4.1. The main methods of data mining technology
Decision tree method. The use of information theory finds the
maximum amount of information in the database, then according
to the different values of the field, a branch of the tree is established.
Tree establishment process is repeated in each sub-set lower nodes
and branches, in order to build decision tree. The most influential
and earliest decision tree algorithm in the world is the ID3 method
developed by Quiulan. The bigger the database, the better it will be.
Since then, a variety of decision tree methods have been developed,
such as the IBLE method, which improves the recognition rate by
10%.
Neural network method. It simulates the structure of human
brain neurons. Based on the MP model and the Hebb learning rule,
the weights are expressed by the weights of neural networks, and
their learning process is reflected in the progressive calculation of
weights of neural networks. The feed network is represented by
perceptron, back propagation model and functional network, and
can be used in prediction and pattern recognition. The feedback
network is represented by discrete models and continuous models
of Hopfield, which are used for associative memory and optimiza-
tion calculations. The self-organizing network is represented by the
ART model and the Koholon model, and used for clustering.
Rough set method. In a database, row elements are treated as
objects, column elements as attributes (conditional attributes and
decision attributes). The equivalence relation is defined as the value
of different objects on an attribute, and the set of objects that sat-
isfy the equivalence relation is called the equivalence class of the
equivalence relation.
Concept tree method. The attribute fields recorded in the
database are abstracted by classification, and the hierarchical struc-
ture is established as a concept tree. Using the concept tree lifting
method can greatly enrich the records in the database. Promotion
of the concept tree for multiple attribute fields results in a highly
generalized knowledge base table, which can then be converted
into rules.
Fuzzy set method. The fuzzy set theory is used to carry out
fuzzy judgment, fuzzy decision, fuzzy pattern recognition and fuzzy
cluster analysis. Fuzziness is objective existence. The higher the
complexity of the system, the lower the accuracy ability, i.e., the
stronger the fuzziness. This is the principle of mutual coherence
summed up by Zadeh.
Visualization technology. Visual data analysis technology
broadens the traditional chart function, and enables users to ana-
lyze data more clearly. For example, the multidimensional data in
a database changes into a variety of graphics, which plays a signif-
icant role in revealing the status, intrinsic nature, and regularity of
data.
4.2. Bayesian algorithm
Bias algorithm is a classification algorithm for predictive mod-
eling. The algorithm calculates conditional probabilities between
input columns and predictable columns based on the assump-
tion that the input attributes are not correlated with each other.
Compared with other algorithms, the algorithm requires a small
amount of computation, so that the mining model can be quickly
generated to discover the relation between the input column and
the predictable column. The algorithm can be used for initial data
detection, and then other algorithms can be used to create other
mining models based on the results of the algorithm.
4.3. Timing algorithm
A sequential algorithm is a regression algorithm used to create
data mining models to predict continuous columns, such as product
sales in a prediction scheme. Other algorithms create prediction
models are based on a given input column, such as a decision tree
model, while the prediction of the timing model is based solely
on the trend derived from the original data set when the model is
created.
An important function of the sequential algorithm is that it can
perform cross prediction. If two separate but related sequences are
used to shape the algorithm, the resulting model can be used to
predict the outcome of a sequence based on the behavior of other
sequences. For example, the actual sales of a product may affect the
forecasted sales of another product.
4.4. Clustering algorithm
Clustering algorithm is a segmentation algorithm. The algorithm
uses iterative techniques to cluster instances of data sets into cate-
gories that contain similar features. These groups are useful when
browsing data, identifying exceptions in the data, and creating pre-
dictions. The cluster analysis model identifies logical relationships
 J. Zhang et al. / Sustainable Computing: Informatics and Systems 20 (2018) 192–202 199

that the data set may not be able to logically observe by random
observation. For example, it is logical to know that people who com-
mute to work usually live not far from their place of work. But the
algorithm can identify other features that are not obvious about
cyclists.
The clustering analysis algorithm is different from other data
mining algorithms. The algorithm can generate clustering analy-
sis model without specifying predictable columns. The clustering
algorithm is strictly based on the data and the relations among the
categories are identified by the algorithms. Clustering algorithm
first identifies the relation of data sets, and generates a series of clas-
sification according to these relations. Scatter diagrams are a very
useful way of visually representing how algorithms group data. A
scatter diagram can represent all instances in a dataset, where each
instance is a point (Fig. 5).

4.5. Association algorithm

Fig. 5. Clustering algorithm.

The association model is generated based on the data set con-

taining the identifier of each instance and the identifier of the
entries contained in each instance. A set of items in an instance
is called an item set. An association model consists of a set of items
in an instance and rules that describe how these items are grouped.
The rules of the algorithm identification can be used to predict
the possible future purchase of the product according to the items
which are already in the customer’s shopping cart.
4.6. Neural network algorithm
The neural network algorithm uses a multilayer perceptron net-
work consisting of three layers of neurons (perceptron), which is
also called the back-propagation Delta rule network. These layers
are the input layer, the optional hidden layer, and the output layer.
In a multilayer perceptron network, each neuron receives one or
more inputs and produces one or more of the same output. Each
output is a simple nonlinear function of the sum of inputs to the
neuron. The input value is passed from the node in the input layer
to the node in the hidden layer and finally to the output layer. There
is no connection between neurons in the same layer.
A mining model constructed using a neural network algo-
rithm can contain multiple networks, depending on the number
of columns used for input and prediction, or depending on the
number of columns being used for prediction only. The number
of networks contains a mining model depended on the number of
states in which the input column and the predicted column are used
by the mining model.

Fig. 6. Structure of neural network.

200 J. Zhang et al. / Sustainable Computing: Informatics and Systems 20 (2018) 192–202
Fig. 7. Regression algorithm instance.
Input neurons provide input attribute values for data mining
models. For discrete input attributes, the input neuron usually rep-
resents a single state of the input property, including the missing
value. For example, a binary input property generates an input node
that indicates the missing or existing state and indicates whether
the property has a value. The Boolean value, which is used as the
input attribute, generates 3 input neurons: one neuron for True
value, one neuron for False value, and one neuron for missing
or existing states. A discrete input property with more than two
states can generate an input neuron for each state and generate an
input neuron for the missing or existing state. A continuous input
attribute can generate two input neurons.
The input neuron can provide input to one or more hidden neu-
rons. The hidden neuron receives input from the input neuron and
provides output to the output neuron.
The output neuron represents the predictable property value of
the data mining model. For discrete input attributes, the output
neuron typically represents a single prediction state of predictable
properties, including missing values. For example, a binary pre-
dictable attribute can generate an output node that describes the
missing or existing state to indicate whether the property has a
value. The applicable range of neural network algorithm is regres-
sion task and classification task (Fig. 6).
4.7. Regression algorithm
The linear regression algorithm is a variant of the decision tree
algorithm, in which the parameter settings are greater than or equal
to the total number of instances of the algorithm. By setting this
parameter in this way, the algorithm will never create a split, thus
performing a linear regression. Linear regression can be used to
determine the relation between two consecutive columns, and the
expression used in this relation is the formula that best represents
the lines of data sequences (Fig. 7).
4.8. Basic concepts of association rules
Using association rules mining, we can get useful information
which has potential value in the mass data. The goal of association
rules is to extract the most interesting patterns in an efficient way.
So far, many efficient association rules have been proposed, and
the Apriori algorithm proposed by Agawal is the most famous one.
Most algorithms are based on the Apriori algorithm, but Apriori
algorithm is facing the challenge, the researchers explored many
new mining methods, and developed the concept of association
rules and application scope.
Assume a data set denoted as I = {i1 , ..., im }, where ik represents
items. A data set is a collection of many transactions.Association
rules can be described as:
A ⇒ B, A ⊂ I, B ⊂ I (15)
The support of an item set is expressed as:
sup(X)
s(X) = p(X) = (16)
|D|
The confidence degree of an item set is expressed as:
sup(X ∪ Y )
s(X) = p(X|Y ) = (17)
sup |D|
The support of association rules represents the importance of
the entire database, while the confidence level reflects its reliability.
Only the association rules with higher support and confidence are
interesting and useful association rules.
According to the completeness of mining patterns, association
rules can be divided into closed frequent itemsets, mining frequent
itemsets, maximal frequent itemsets and constrained frequent
itemsets. The association rules can be divided into single layer
association rules, multi-level association rules, single dimension
association rules and multidimensional association rules accord-
ing to the layers and dimensions of the data involved in the rule.
According to the types of values processed by the rules, associ-
ation rules can be divided into mining Boolean association rules
and quantitative association rules. According to the types of min-
ing rules, association rules can be divided into association rules and
related rules mining. According to the mining pattern types, associ-
ation rules can be divided into frequent itemsets mining, sequential
pattern mining, structural pattern mining, and so on. According to
the types of mining constraints, association rules can be divided
into knowledge type constraints, data constraints, interestingness
constraints, and rules constraints.
4.9. A mining algorithm of association rules
At present, the mining algorithms of association rules can be
broadly divided into breadth first algorithm, depth first algorithm,
data set partitioning algorithm, sampling algorithm, incremental
updating algorithm and so on. Breadth first algorithm is also called
hierarchical algorithm, including Apriori, AprioriTid and AprioriHy-
brid algorithms proposed by Agrawal et al. DHP algorithm proposed
by Park et al.
Apriori algorithm is also a breadth first algorithm, AprioriTid
algorithm is based on the Apriori algorithm evolved. The Apriori
algorithm uses the algorithm first scans the database again, when
the scanar is no longer scanning the entire database, it also cal-
culates the frequent itemsets support, to improve the efficiency
of the algorithm and to reduce the time of scanning the database.
Convergence of Apriori algorithm and AprioriTid algorithm in the
AprioriHybrid algorithm, Apriori algorithm using the initial scan of
the database, when generating candidate set size can be stored into
memory for processing and then turned to the AprioriTid algorithm,
to find all the frequent item sets. DHP algorithm uses hash table
technology to prune data sets and candidate itemsets to reduce the
time and space cost.
In depth first algorithms, the common algorithms are FP-growth
algorithm, OP algorithm, TreeProjection algorithm and so on. The
FP-growth algorithm only scans the database two times and avoids
scanning the database several times. FP-growth does not need to
generate a large set of candidate items, which greatly reduces the
search space in the mining process, and has an order of magnitude
improvement in both time efficiency and space efficiency. But its
 J. Zhang et al. / Sustainable Computing: Informatics and Systems 20 (2018) 192–202
application difficulty is that when dealing with large and sparse
databases, considerable space is needed in mining, processing, and
recursive computation.
4.10. Data set partitioning algorithm
The data set partitioning algorithm includes the Partition algo-
rithm proposed by Savasere and the DIC algorithm proposed by
Brin. Partition algorithm is a data block of several independent
and can be stored in memory process, and save disk access over-
head. It considers each logical block individually to generate the
corresponding frequency set, then generates all possible global
candidate itemsets using frequent itemsets, and finally scans the
database again to calculate the support of the itemsets and per-
forms global enumeration. The whole process requires only two
scans of the database, but the number of candidate itemsets is large.
The DIC algorithm also takes database partition of the database,
which will be divided into several partitions at the beginning of
each part of the mark. In the process of scanning the database, it can
be added in the candidate itemsets in each partition of the mark-
ing point, parallel computing support may be set in the calculation
of frequency itemsets. Algorithm of scanning database basically is
less than the maximum number of frequency sets. When the data
blocks are properly partitioned, all the frequent itemsets can be
found by only scanning the database two times.
In the algorithm based on partitioning, the main bottleneck is
the time of algorithm execution, and the accuracy of the frequent
itemsets is not very high. However, the algorithm of this type has
high parallelism. It only needs to scan the database two times,
which greatly reduces the I/O operation, thus improving the effi-
ciency of the algorithm.
4.11. Incremental updating algorithm
The incremental updating algorithm uses the mining association
rules to discover new association rules on the changed database or
parameters, and deletes the outdated association rules to maintain
the updating problem of data sets. At present, most of the incremen-
tal updating algorithms are improved and evolved with the Apriori
algorithm, including the FUP algorithm proposed by D.W.Cheung,
IUA, PIUA, IUAR algorithm and so on.
FUP algorithm is the improvement of Apriori algorithm, and
it is also a classical algorithm to solve the incremental updat-
ing problem. The FUP algorithm mainly aims at how to generate
the association rules of the updated database when the database
is modified under the condition that the minimum support and
the minimum confidence level remain unchanged. It uses the fre-
quent itemsets information obtained by mining process to avoid
the repeated computation of the time cost of the support number
of frequent itemsets and to improve the efficiency of the algorithm.
4.12. Parallel mining algorithm
The parallel algorithm uses the set of simultaneous processes,
interactions and coordination to complete the solution of a given
problem. Including CD, DD and CaD algorithms proposed by
Agrawal et al. PDM algorithm proposed by Park et al. DMA and
FDM algorithms proposed by Cheung et al.
The CD algorithm allows parallel redundant computations on
idle processors to reduce traffic, and speeds almost linearly to
speedup. But its disadvantage is that both traffic and candidate
frequent itemsets are relatively large.
DD algorithm for the candidate set is divided into each proces-
sor to overcome the shortcomings of CD algorithm, DD algorithm,
however due to the data of mobile solutions of low efficiency in
communication load, large interactive mode between processors
201
is easy to cause the processor is in an idle state, and each trans-
action records are based on multiple hash tree processing leads to
redundant computation.
The CaD algorithm attempts to reduce the data dependency
between processors by partitioning databases and candidate sets
so that each processor can compute independently.
The PDM algorithm is similar to the CD algorithm, all processors
have the same hash table and candidate set. The process of generat-
ing parallel candidates is to generate a set of candidate sub itemsets
through each processor, and then exchange all the sub itemsets on
all the processors to generate a global candidate set. But the PDM
algorithm involves a large number of disk I/O operations.
5. Conclusion
Intelligent system integration mainly includes artificial intel-
ligence, computational intelligence method and other intelligent
technology. We introduce the application and technology of several
intelligent system integration, and the advantages and disadvan-
tages of learning theory and expert system, plus, neural network is
applied in intelligent systems. When solving the intelligent com-
putation, the solution of the problem is not needed, or the gradient
information of the system is not required, so the continuous and
discrete problems can be dealt with independently. Intelligent
computing method can figure out the optimal solution of the global
optimization problem with a greater probability for different opti-
mization problems, and the intelligent calculation method can
be introduced easily with the heuristic rules of logic calculation
method which is simple and easy to understand.
In this paper, the mining of association rules in data mining is
discussed carefully and clearly, and some commonly used min-
ing algorithms are analyzed, compared and summarized based
on statistics. The existing improved algorithms can not meet the
needs of people’s fast and timely response to the mining system.
Therefore, we need to improve the efficiency of the mining pro-
cess and interact with the user to generate the visual results. At the
same time, we introduce various techniques and models of pattern
recognition. Several examples show that the intelligent computing
system based on pattern recognition and data mining algorithm
has higher efficiency and recognition rate.
Acknowledgement
This paper is supported by the Large-scale public building
energy consumption data fusion platform (Shaanxi Provincial
Department of Education grant projects, Shaanxi [2016] 250 docu-
ment)
References
[1] Poonam Sinai Kenkre, Anusha Pai, Louella Colaco, Real time intrusion
detection and prevention system, in: Proceedings of the 3rd International
Conference on Frontiers of Intelligent Computing: Theory and Applications
(FICTA) 2014, Springer, Cham, 2015.
[2] Fei Tao, et al., CCIoT-CMfg: cloud computing and internet of things-based
cloud manufacturing service system, IEEE Trans. Ind. Inf. 10 (2) (2014)
1435–1442.
[3] H. Wang, J. Wang, An effective image representation method using kernel
classification, in: 2014 IEEE 26th International Conference on Tools with
Artificial Intelligence (ICTAI), IEEE, November, 2014, pp. 853–858.
[4] Ahmed Patel, et al., An intrusion detection and prevention system in cloud
computing: a systematic review, J. Netw. Comput. Appl. 36 (1) (2013) 25–41.
[5] Batya Friedman, Peter H. Kahn, Human agency and responsible computing:
Implications for computer system design, J. Sys. Software 17 (1) (1992) 7–14.
[6] C. Bi, H. Wang, R. Bao, November. SAR image change detection using
regularized dictionary learning and fuzzy clustering, in: 2014 IEEE 3rd
International Conference on Cloud Computing and Intelligence Systems
(CCIS), IEEE, 2014, pp. 327–330.
202 J. Zhang et al. / Sustainable Computing: Informatics and Systems 20 (2018) 192–202
[7] T. Qiu, Y. Zhang, D. Qiao, X. Zhang, M.L. Wymore, A.K. Sangaiah, A Robust Time
Synchronization Scheme for Industrial Internet of Things, IEEE Trans. Ind. Inf.
(2017), http://ieeexplore.ieee.org/abstract/document/8008773/ (in press).
[8] Pachipala Yellamma, Narasimham Challa, V. Sreenivas, Intelligent Data
Security In Cloud Computing, Int. J. Curr. Eng. Technol. 4 (1) (2014).
[9] S. Zhang, H. Wang, W. Huang, Two-stage plant species recognition by local
mean clustering and Weighted sparse representation classification, Cluster
Comput. (2017) 1–9.
[10] D.V. Medhane, A.K. Sangaiah, ESCAPE: Effective Scalable Clustering Approach
for Parallel Execution of continuous position-based queries in position
monitoring applications, IEEE Trans. Sustain. Comput. 2 (2) (2017) 49–61,
http://dx.doi.org/10.1109/TSUSC.2017.2690378.
[11] Jinsung Byun, et al., Intelligent household LED lighting system considering
energy efficiency and user satisfaction, IEEE Trans. Consumer Electron. 59 (1)
(2013) 70–76.
[12] A.K. Sangaiah, O.W. Samuel, X. Li, M. Abdel-Basset, H. Wang, Towards an
efficient risk assessment in software projects-fuzzy reinforcement paradigm,
Comput. Electr. Eng. (2017), http://www.sciencedirect.com/science/article/
pii/S0045790617305530 (in press).
[13] Oscar Castillo, Patricia Melin, Janusz Kacprzyk (Eds.), Recent advances on
hybrid intelligent systems, Springer, 2013.
[14] Omid Fatahi Valilai, Mahmoud Houshmand, A collaborative and integrated
platform to support distributed manufacturing system using a
service-oriented approach based on cloud computing paradigm, Robot.
Comput.-Integrat. Manuf. 29 (1) (2013) 110–127.
[15] M. Zareapoor, P. Shamsolmoali, J. Yang, Kernelized support vector machine
with deep learning: an efficient approach for extreme multiclass dataset,
Pattern Recognit. Lett. (2017), http://www.sciencedirect.com/science/article/
pii/S0167865517303276 (in press).
[16] Neeraj Kumar, Jong-Hyouk Lee, Joel J.P.C. Rodrigues, Intelligent mobile video
surveillance system as a Bayesian coalition game in vehicular sensor
networks: learning automata approach, IEEE Trans. Intell. Transp. Syst. 16 (3)
(2015) 1148–1161.
[17] Sukhpal Singh, Inderveer Chana, A survey on resource scheduling in cloud
computing: Issues and challenges, J. Grid Comput. 14 (2) (2016) 217–264.
[18] B.B. Gupta, S. Gupta, P. Chaudhary, Enhancing the browser-side context-aware
sanitization of suspicious html5 code for halting the dom-based XSS
vulnerabilities in cloud, Int. J. Cloud Appl. Comput. (IJCAC) 7 (1) (2017) 1–31.
[19] S. Bakshi, P.K. Sa, H. Wang, S.S. Barpanda, B. Majhi, Fast periocular
authentication in handheld devices with reduced phase intensive local
pattern, Multimedia Tools Appl. (2017) 1–29.
[20] J. Jin, J. Gubbi, S. Marusic, M. Palaniswami, An information framework for
creating a smart city through internet of things, IEEE Internet Things J. 1 (2)
(2014) 112–121.
[21] Chen Chen, Xiaomin Liu, Tie Qiu, Lei Liu, Arun Kumar Sangaiah, Latency
estimation based on traffic density for video streaming in the internet of
vehicles Computer Communications, 111, 2017, pp. 176–186, http://dx.doi.
org/10.1016/j.comcom.2017.08.010, ISSN 0140-3664.
[22] R. Nasim, A.J. Kassler, A. Antonic, Mobile publish/subscribe system for
intelligent transport systems over a cloud environment, in: 2014 International
Conference on Cloud and Autonomic Computing (ICCAC), IEEE, 2014.
[23] Sundarapandian Vaidyanathan, A novel chemical chaotic reactor system and
its output regulation via integral sliding mode control. parameters 1 (2015): 4.
[24] A.K. Sangaiah, A.K. Thangavelu, X.Z. Gao, N. Anbazhagan, M.S. Durai, An ANFIS
approach for evaluation of team-level service climate in GSD projects using
Taguchi-genetic learning algorithm, Appl. Soft Comput. 30 (2015) 628–635.
[25] A.K. Sangaiah, A.K. Thangavelu, An adaptive neuro-fuzzy approach to
evaluation of team-level service climate in GSD projects, Neural Comput.
Appl. 25 (3–4) (2014) 573–583.
[26] J. Wang, Y. Zhou, H. Wang, X. Yang, F. Yang, A. Peterson, Image tag completion
by local learning, in: International Symposium on Neural Networks, Springer,
Cham, October, 2015, pp. 232–239.
[28] V. Jain, A.K. Sangaiah, S. Sakhuja, N. Thoduka, R. Aggarwal, Supplier selection
using fuzzy AHP and TOPSIS: a case study in the Indian automotive industry,
Neural Comput. Appl. (2016) 1–10.