J. Software Engineering & Applications, 2010, 3, 588-592
doi:10.4236/jsea.2010.36068 Published Online June 2010 (http://www.SciRP.org/journal/jsea)
A Neuro-Fuzzy Model for QoS Based Selection of
Web Service
Abdallah Missaoui1, Kamel Barkaoui2
1LSTS-ENIT, Tunis, Tunisia; 2CEDRIC-CNAM, Paris, France.
Received January 5th, 2010; revised March 8th, 2010; accepted March 10th, 2010.
ABSTRACT
The automatic selection and composition of Web services rely strongly on the manner to deal with ambiguity inherent to
the description of functionalities of these services and the client’s requests. Quality of Service (QoS) criteria become
crucial in Web services selection and the problem of checking that a web service satisfies a given level of QOS is
considered in recent research works. This paper presents a QoS based automatic classification method of web services.
These services give generally similar functionalities and are offered by different providers. The main feature of our Web
service selection model is to take advantage of the neuro-fuzzy logic for coping with the imprecision of QoS constraints
values.
Keywords: Web Service, Selection, Neuro-Fuzzy, QoS, Constraint
1. Introduction
Web services are modular, self-contained, self-describing
software components which are distributed over the Web.
They can be readily located and checked-out online and
dynamically, using a new directory and corresponding
search mechanism known as Universal Description, Dis-
covery, and Integration (UDDI).
The requester accesses the description using a UDDI or
other types of registry, and requests the execution of the
provider’s service by sending a SOAP message to it (see
Figure 1).
SOAP and HTTP provide exactly what they were de-
signed for a simple, lightweight mechanism for interop-
Figure 1. Basic web services architecture
erability and distributed communication. However, SOAP
and HTTP do not provide the traditional enterprise quali-
ties of service typically needed for an enterprise.
Furthermore, SOAP was designed to be extensible,
and it can be extended to support any desired QoS fea-
ture by adding SOAP headers to the SOAP messages and
adding QoS features to the basic SOAP run-time facili-
ties.
In recent years, several service providers offer QoS
features to there customers. Then, multiple providers
may provide similar functionalities with different values
of non-functional properties.
Their non-functional properties need to be considered
during service selection. There are characterised as qual-
ity of service (QoS). In many practice cases of business
applications, it is recommended to be taken into account
during the provider selection.
The human faculty of cognition and perception is very
complex, but it possesses an efficient mechanism for in-
formation processing and expression [1,2].
This paper applies the neuro-fuzzy decision making
approach in the process of selection and choice of the
most appropriate web service with respect to quality of
service criteria.
This paper is organized as follows: Section 2 presents
web service QoS generic description. In Section 3, we
discuss and evaluate related works on web service selec-
tion adopting a common fuzzy logic approach. Section 4,

A Neuro-Fuzzy Model for QoS Based Selection of Web Service589
we enlighten our QoS requirement description model ex-
ploiting neuro-fuzzy logic in order to deal with the im-
precision of QoS constraints values. Comments and
recommendations for our model are explicitly presented
in Section 5. Finally, Section 6 draws a conclusion.
2. QoS Properties of Web Service
Many services are appearing on the Web, several requ-
esters are presented to a group of service providers offer-
ing similar services. Different service providers may have
different qualities of service.
QoS is one of the most important factors for user’s
choice of Web service. This will require sophisticated pa-
tterns of selection process. It is necessary to provide an
appropriate negotiation mechanism between clients and
service providers to reach mutually-agreed QoS goals.
QoS management in Web service architecture includes
the definition of QoS attributes and the specifications of
the following processes: QoS publication, discovery,
validation, and monitoring. Many works have studied
QoS management on web service. Several QoS languages
and architectures are proposed.
The proposed approaches for QoS management can be
classified into two groups: one based on extending web
service technologies including SOAP, WSDL and UDDI
to support QoS [3-5]. The second group use independent
entities to perform QoS management [6].
Quality of service is defined by the ability to provide
different priorities to different applications, users, or data
flow, or to guarantee a certain level of performance to a
data flow. A QoS property may include several sub-pro-
perties representing different evaluation criteria, e.g. avai-
lability, performance, accessibility. In addition, a QoS
property can be evaluated by many metrics and therefore
it is necessary to define the units of measurements.
QoS in web service architecture is a combination of
several qualities or properties of a service, such as:
 Response time: the interval between a user-
command and the reception of an action, a result or a
feedback from the service.
 Availability: availability is the percentage of
time that a service is available for use;
 Accessibility: Accessibility represents the de-
gree that a system is normatively operated to coun-
teract request messages without delay.
 Throughput: It means the max number of ser-
vices that a platform providing Web services can
process for a unit time.
 Reliability: Reliability is the quality aspect of a
Web service that represents the degree of being ca-
pable of maintaining the service and service quality.
The number of failures per month or year represents
a measure of reliability of a Web service.
 Price: represents the money that the customer
should pay for this service. It is always associated
with the value of the service’s functionality, i.e. the
more a service costs, the more complicated functions
it provides.
 Security Level: represents the security level of
a service. It includes the existence and type of au-
thentication mechanisms the service offers, confi-
dentiality and data integrity of messages exchanged,
non-repudiation of requests or messages, and resil-
ience to denial-of-service attacks [7].
3. Related Work
With the strong popularity of the development of service
oriented application, quality of service becomes a central
interest of more and more researchers and enterprises. QoS
values are proportional to the reliability degree and per-
formance of service and thus play a very important role in
the provider choice. A large number of services are ex-
posed constraint information’s for comparison providers.
Many researches [5,8-10] have studied QoS issues to
improve two processes of discovery and selection of ser-
vices. Several QoS-aware web service selection mecha-
nisms have been developed in recent years in order to
perform the web service composition and to improve
performance applications based on services. This mecha-
nisms’ main objective is how to how properly select a set
of providers that most satisfy the constraints defined by
the user in his business processes.
Menascé studies the problem of finding services that
minimize the total execution time. It presents an opti-
mized heuristic algorithm that finds the optimal solution
without exploring the entire solution space. The solution
provided in [11] covers only the important case of execu-
tion constraints but not all QoS properties.
Pfeffer proposed a fuzzy logic based model for repre-
senting any kind of non-functional service properties. This
representation of user preferences enables the fast eva-
luation of the requested service composition by a fuzzy
multiplication of the service composition properties. Thus
service composition’ properties are measured during or
after execution [12].
Other works have been done in fuzzy logic based web
service selection. In [12-17], various methods have been
proposed for specifying fuzzy QoS constraints and for
ranking Web services based on their fuzzy representation.
There is a more suitable technique to quantify func-
tional properties: Linear Programming. These properties
are not fitting well for measuring the non-functional at-
tributes, because the majority of them are not easy to be
quantified in numerical form. In the meantime, user’s QoS
constraints regularly remain imprecise or ambiguous due
to various human mental states, and it is very difficult to
distinguish the priority order among QoS criteria.
Furthermore, in web services selection, the applied QoS
constraints are not explicitly defined. It is necessary to
relax the constraints to make an optimal solution. The use

A Neuro-Fuzzy Model for QoS Based Selection of Web Service 590
of fuzzy logic offers improvements in the overall satis-
faction level. The QoS information’s represented at ab-
stract level such that it could efficiently select the best
services.
However they are still initial efforts which need further
investigation for more complete solutions. In the follow-
ing, we specify several open issues that can be solved:
 When we use some kinds of fuzzy numbers like
triangular fuzzy they may not be easy to be defined
by end users.
 It is very important to correctly define the QoS
properties that we use in the selection process. These
criteria’s QoS have important effects on ranking
methods.
 How to improve fuzzy based web service dis-
covery and the representation of QoS to achieve ef-
fective web service selection?
 How to automatically set the weights of service
providers attributes?
4. Refinement of the Framework
Neuro-fuzzy technique is the combination of two artificial
intelligence (AI) methods: fuzzy logic techniques and
neural networks. Neuro-fuzzy system has the ability to
handle the nonlinear and complex systems. It is con-
structed based on the learning algorithm of neural net-
works technique to adjust the appropriate parameters for
fuzzy logic system [18].
In this paper, we aim to solve the selection of web ser-
vices in a global and flexible manner by introducing a
neuro-fuzzy way. For this purpose, we have developed a
neural-fuzzy system based on the Sugeno Approach [19].
This is known as the ANFIS (i.e., Adaptive Neuro-Fuzzy
Inference Systems). We assume that semantic match-
making has taken place to identify functionally equivalent
services. When several of them are available to perform
the same task, their quality aspects become important and
useful in the selection process.
An ANFIS is a multi-layered feed forward network, in
which each layer performs a particular task. The layers are
characterized by the fuzzy operations they perform. Fig-
ure 2 describes the global architecture of this neural-fuzzy
system. It shows a n-input, type-5 ANFIS. Three member-
ship functions are associated with each input.
We assume that the fuzzy inference system under con-
sideration has n inputs Q1, Q2,…,Qn (which are one service
attributes). Each input has five linguistic terms, for ex-
ample, the input Q1 possesses the terms {M11, M12,…, M15}.
For each input Qi, we have defined linguistic expres-
sions
Li = {Very Poor(vp), Poor(p), Medium(m), Good(g), Very
Good(vg)}
The common fuzzy if-then rule has the following type:
Rule 1: If (Q1 is M11) and (Q2 is M21) and … and (Qn is Mn1)
then f1 (Q1, Q2,…, Qn)
We denote the output of the ith node in layer k as Ok,i.
Figure 2 shows the schematic diagram of the ANFIS
structure, which consists of five layers.
Layer 1: Every node i in this layer transform the crisp
values to a fuzzy one
1, 1
1()
iM
i
OQ


for


1, 2,...,5i and
1, 2
2()
iM
i
OQ


for


1, 2,...,5i and, …, and
1, () iMn
ni
OQ


for


1, 2,...,5i
where QK is the input to node K and Mki (and

1,...,kn



1,...,5i) is a linguistic label (very poor, poor, fair,
good, very good) associated with this node. In other words,
O1,i is the membership grade of a fuzzy set




1 5
,...,
nn
M M

11,...MM 1521 25
,,..., ...MM M
and it specifies the degree which the given input QK
(


1,...,kn) satisfies the quantifier M.
We use the following generalized Bell function as the
membership function (MF)
2
1
()
1
bi
M
i
i
Q
Qc
a
 


where ai, bi and ci are the parameters set of MF. The
bell-shaped function varies depending on the values of
these parameters. Where the parameters a and b vary the
width of the curve and the parameter c locates the center




Figure 2. The structure of the neural fuzzy selector

A Neuro-Fuzzy Model for QoS Based Selection of Web Service591
of the curve. The parameter b should be positive. The pa-
rameters in this layer are referred to as premise para-
meters. The generalized Bell-shaped function is shown in
Figure 3.
Layer 2: Every node in this layer is a fixed node labeled
∏. The weighting factor, wk, of each rule is calculated by
evaluating the membership expressions in the antecedent
of the rule. This is accomplished by first converting the
input values to fuzzy membership values by utilizing the
input membership functions and then applying the and
operator to these membership values.
The and operator corresponds to the multiplication of
input membership values.
2,1 2
12
( )()...()
iiM MMn
ii ni
OwQ QQ

 
Each node output represents the firing strength of a
rule.
Layer 3: Every node in this layer is a fixed node labeled
N. The function of the fixed node is used to normalize the
input firing strengths.
3,
1
i
ii
n
j
j
w
Ow
w




1,...,in
Layer 4: Every node in Layer 4 is a parameterized
function, and the adaptive parameters are called “conse-
quent parameters”.
The node function is given by:
4,1 12 111
(... )
iii i
iiiin n
OwfwpQpQpQp

  
Layer 5: The single node in this layer is a fixed node
labeled ∑, which computes the overall output as the
summation of all inputs:
1
5,1
1
1
n
i
n
i
in
i
i
i
wf
Owf
w







Thus, the ANFIS network is constructed according to
the TSK fuzzy model. This ANFIS architecture can then
update its parameters according to the backpropagation
algorithm [20].
This algorithm minimizes the error between the output
Figure 3. Generalized bell-shaped (a = 2, b = 4, c = 6)
of the ANFIS and the desired output.
Our neuro-fuzzy system allows classifying service
providers in several categories: very poor, poor, fair,
good, very good. It allows automating the selection pro-
cess in the dynamic composition of services.
According to the QoS requirements of web service
providers and the functions of Neuro-fuzzy system, we
believe that each service invoked is appropriate candidate
to increase the composition ability of web services and to
decrease the burden of composition cognition and the
minimal development cost.
5. Comments and Recommendations
In fuzzy inference system (FIS), The MF of the conse-
quent of each rule is a constant of a fuzzy MF. There are
two steps to construct this system: the specification of an
appropriate number of input/output and the specification
of the shape of MFs. The main problem is that structure
identification requires human expertise to solve the pa-
rameter estimation. In our selector system we used a dif-
ferent approach, which take advantage of adaptive neural
networks algorithms during fitting procedures. MF pa-
rameters are fitted to a dataset through a learning algo-
rithm.
A significant number of samples of service providers
are needed in order to have better result and to avoid
having too many defect values during selection process.
The database must be as complete as possible, including
samples of providers attributes over a broad range. The
number of samples depends on the context and on the
runtime environment.
On the other hand, fuzzy logic sets are based on trans-
parence, linguistic rules and establish a framework to
include human expertise into modelling. The number of
rules is decided by an expert who is familiar with the
system to be modeled. In our work, however, no expert is
available and the number of membership functions as-
signed to each input qualities is chosen empirically by
examining the desired input-output data.
We merged the fuzzy logic approach with the ability of
learning algorithms from neural networks to adjust the
model.
6. Conclusions
Web service composition is an emerging area involving
important technological challenges. Some of the main
challenges are to correctly describe QoS of Web services,
to compose them adequately and automatically, and to
discover suitable providers and QoS composition issues.
Neuro-fuzzy logic can be seen as a promising formal
technique for representing imprecise QoS constraints. In
this paper, we have presented a solution to use neuro-
fuzzy approach in Web service discovery and selection.
We have proposed methods for ranking and selecting web
services based on a neuro-fuzzy specification of fuzzy

A Neuro-Fuzzy Model for QoS Based Selection of Web Service 592
QoS constraints. The user’s constraints are formalized as
fuzzy sets and the Qos criteria’s are expressed as fuzzy
expressions.
This model can be seen as a contribution towards a
more complete solution for web service composition in-
tegrating fully QoS features.
REFERENCES
[1] J. S. Jang, “ANFIS: Adaptive-Network-Based Fuzzy In-
ference System,” IEEE Transactions on Systems, Man,
and Cybernetics, Vol. 23, No. 3, 1993, pp. 665-684.
[2] J. R. Jang and C. T. Sun, “Neuro-Fuzzy and Soft Com-
puting: A Computational Approach to Learning and
Machine Intelligence,” Prentice-Hall, Inc., Upper Saddle
River, New Jersy, 1997.
[3] V. Diamadopoulou, C. Makris, Y. Panagis and E. Sakko-
poulos, “Techniques to Support Web Service Selection
and Consumption with QoS Characteristics,” Journal of
Network and Computer Applications, Vol. 31, No. 2, 2008,
pp. 108-130.
[4] A. F. M. Huang, C. W. Lan and S. J. H. Yang, “An Optimal
QoS-Based Web Service Selection Scheme,” Information
Sciences, Vol. 179, No. 19, 2009, pp. 3309-3322.
[5] L. Zeng, B. Benatallah, A. H. H. Ngu, M. Dumas, J.
Kalagnanam and H. Chang, “QoS-Aware Middleware for
Web Services Composition,” IEEE Transactions on
Software Engineering, 2004, pp. 311-327.
[6] D. A. Menascé, H. Ruan and H. Gomaa, “QoS Manage-
ment in Service-Oriented Architectures,” Journal of Per-
formance Evaluation, Vol. 64, No. 7-8, 2007, pp. 646-663.
[7] D. A. Menasce, “QoS Issues in Web Services,” IEEE
Internet Computing, Vol. 6, No. 6, 2002, pp. 72-75.
[8] M. Sultana, M. M. Akbar and M. Rouf, “Network Flow
Heuristic Algorithm for a Distributed Web Service Selection
Problem,” IEEE Conference on Communications, Compu-
ters and Signal Processing, 2009, pp. 465-470.
[9] D. Tsesmetzis, I. Roussaki and E. Sykas, “QoS-Aware
Service Evaluation and Selection,” European Journal of
Operational Research, Vol. 191, No. 3, 2008, pp. 1101-
1112.
[10] S. Chaari, Y. Badr and F. Biennier, “Enhancing Web Ser-
vice Selection by QOS-Based Ontology and WS-Policy,”
Proceeding of the 23rd ACM Symposium on Applied
Computing, Ceará, 2008, pp. 2426-2431.
[11] D. A. Menascé, E. Casalicchio and V. Dubey, “On
Optimal Service Selection in Service Oriented Archi-
tectures,” Performance Evaluation Journal, Vol. 67, No. 8,
2010, pp. 659-675.
[12] H. Pfeffer, S. Krüssel and S. Steglich, “A Fuzzy Logic
based Model for Representing and Evaluating Service
Composition Properties,” The Third International Con-
ference on Systems and Networks Communications,
Bangalore, 2009.
[13] M. Lin, J. Xie, H. Guo and H. Wang, “Solving Qos-Driven
Web Service Dynamic Composition as Fuzzy Constraint
Satisfaction,” IEEE International Conference on e-Tech-
nology, e-Commerce and e-Service, Hong Kong, 2005.
[14] P. Wang, K. Chao, C. Lo, C. Huang and Y. Li, “A Fuzzy
Model for Selection of QoS-Aware Web Services,” IEEE
International Conference on e-Business Engineering,
IEEE Computer Society, Shanghai, 2006, pp. 585-593.
[15] K. M. Chao, M. Younas, C. C. Lo and T. H. Tan, “Fuzzy
Atchmaking for Web Services,” The 19th International
Conference on Advanced Information Networking and
Applications, Taipei, 2005.
[16] L. Zhuang, Y. F. Huang, W. G. Jian, J. B. Zhou and H. Q.
Guo, “Solving Fuzzy QoS Constraint Satisfaction Tech-
nique for Web Service Selection,” International Con-
ference on Computational Intelligence and Security Work-
shops, Harbin, 2007.
[17] H. Tong and S. Zhang, “A Fuzzy Multi-Attribute Decision
Making Algorithm for Web Services Selection Based on
QoS,” The IEEE Asia-Pacific Conference on Services
Computing, Guangzhou, 2006.
[18] M. A. Denai, F. Palis and A. Zeghbib, “ANFIS Based
Modelling and Control of Non-Linear Systems: A Tu-
torial,” IEEE International Conference on Systems, Man
and Cybernetics, Vol. 4, 2004, pp. 3433-3438.
[19] O. Nelles, A. Fink, R. Babuka and M. Setnes, “Com-
parison of Two Construction Algorithms for Takagi-
Sugeno Fuzzy Models,” International Journal of Applied
Mathematics and Computer Science, 2000, pp. 835-855.
[20] P. Werbos, “The Toots of the Back Propagation: From
Ordered Derivatives to Neural Networks and Political
Forecasting,” John Wiley and Sons, Inc, New York, 1993.