You are on page 1of 8

Tamkang Journal of Science and Engineering, Vol. 9, No 1, pp.

63-70 (2006)

63

Design of PSO-based Fuzzy Classification Systems
Chia-Chong Chen
Department of Electronics Engineering, Wufeng Institute of Technology, Chiayi, Taiwan 621, R.O.C.

Abstract
In this paper, a method based on the particle swarm optimization (PSO) is proposed for pattern classification to select a fuzzy classification system with an appropriate number of fuzzy rules so that the number of incorrectly classified patterns is minimized. In the PSO-based method, each individual in the population is considered to automatically generate a fuzzy classification system for an M-class classification problem. Subsequently, a fitness function is defined to guide the search procedure to select an appropriate fuzzy classification system such that the number of fuzzy rules and the number of incorrectly classified patterns are simultaneously minimized. Finally, two classification problems are utilized to illustrate the effectiveness of the proposed PSO-based approach. Key Words: Particle Swarm Optimization, Fuzzy Classification System, Pattern Classification

1. Introduction
Fuzzy rule-based systems have been successfully applied to solve many classification problems. In many classification problems, fuzzy classification rules are derived from human experts as linguistic knowledge. Because it is not usually easy to derive fuzzy rules from human experts, many approaches have recently been proposed to generate fuzzy rules automatically from the training patterns of the considered classification problem [3-5,7,8,13]. In order to generate fuzzy rules from training patterns, fuzzy partition in the input space are generally considered to determine the premise part of a fuzzy classification system. The grid-type fuzzy partition of the input space [3,4,7,9,11,13] and the scatter-type fuzzy partition of the input data [8,10] have been often used to model fuzzy systems for training patterns. In [3], a heuristic method for generating fuzzy rules is applied to the grid-type fuzzy partitions, and a rule selection method, based on genetic algorithms, is then employed to select the relevant fuzzy rules from the generated fuzzy rules for classifying the training patterns in the considered classification problem. However, this approach to solv*Corresponding author. E-mail: ccchen@mail.wfc.edu.tw

ing classification problems with high-dimensional pattern spaces has a significant shortcoming: the number of fuzzy rules becomes enormous as the number of dimensions increases and the learning time for genetic algorithms is too high. Furthermore, an adaptive grid partition in the input space is used to design the ANFIS-based fuzzy classifier in [4]. This approach takes the uniformly partitioned grid as the initial state. The grid evolves as the parameters in the premise membership functions are adjusted. However, the adaptive grid-partition scheme has two problems. First, the number of fuzzy sets for each input variable is predetermined in advance. Second, the learning complexity suffers from an exponential explosion as the number of inputs increases. In the scatter partition for constructing fuzzy classifiers, fuzzy minmax hyperbox classifiers are powerful tools for solving classification problems. Simpson [8] proposed a method for generating the hyperbox regions to construct a fuzzy classifier to the considered classification problem. In this approach to the classification problem, the learning parameter is very critical, since it directly affects the number and position of the resulting hyperboxes. Consequently, it influences the structure of the fuzzy classifier and the classification performance so that this approach suffers from a high sensitivity of the classification accu-

In the PSO-based approach. the shape of the membership function is determined by a parameter vector mji = [m(ji. ÷ ø if xi £ m( ji . m(ji. m.m( ji . ÷ ø where m(ji. Thus. The goal of this paper is to find an appropriate fuzzy classification system to minimize both of the number of incorrectly classified patterns and the number of fuzzy rules.1) . M}. the fixed learning parameter of the above approach to fuzzy min-max hyperbox classifier makes the same constraint on coverage resolution in the whole input space so that some small hyperboxes are generated even in the input space far from class boundaries.2. m( ji .th rule : Then x = ( x1 . Besides.1) . (2) 2 ö ÷ .3) î è è ö ÷ ÷ ø ö ÷ ÷ ø 2 ö ÷.2) m(ji. the set of parameters in the premise part of the rule base is defined as r = [r1 r2 ¼ rR]. Hence. respectively. ship function of the fuzzy set Aji is described by m Aji (m( ji . each individual in the population is considered to represent a fuzzy classification system. Structure of Fuzzy Classification Systems When an M-class classification problem is considered. each parameter set consisting of the premise and consequent parameters determines a fuzzy classification system.1) ïexp ç .ç i ç m( ji . if xi > m( ji . xm) is given. This reduces the generalization capability of the fuzzy min-max hyperbox classifier. the left and right width values of the membership function. According to (1). different parameter sets determine different fuzzy classification systems so that the generated fuzzy classification systems have different performances. Section 5 draws conclusions about the proposed approach to solving the classification problem.1) . Then. 3.ç ï ç ç m( ji .64 Chia-Chong Chen racy with respect to the skill of the user to determine the appropriate learning parameter. a method based on the particle swarm optimization (PSO).¼. 2.¼. In this paper.1). In the next section. x2. Section 4 considers classification problems of a synthetic data set and the Iris data set to illustrate the learning ability and the generalization ability of the proposed approach. xm ) belongs to Class H j with CF = CFj . When the input x = (x1. Aji. are the premise fuzzy sets of the j-th fuzzy rule. The j-th fuzzy rule in the rule base is determined by a parameter vector rj = [mj1 mj2 ¼ mjM]. Section 2 describes the structure of the fuzzy classification system. i = 1. xi ) ì æ æ x -m ( ji . called a PSO-based method. The rest of this paper is organized as follows. the set of parameters in the consequent part of the rule base is defined as a = [H1 CF1 H2 CF2¼HR CFR]. a fitness function is implemented to guide the search procedure to select an appropriate fuzzy classification system such that the number of fuzzy rules and the number of incorrectly classified patterns are simultaneously minimized.L .1] is the grade of certainty of the j-th fuzzy rule. Hj Î{1.3) determine the center position.¼. Section 3 proposes a PSO-based method to select an appropriate fuzzy classification system for the considered classification problem.3)].L .1) ïexp ç . and CFj Î [0.3) . R. the firing strength of the premise of the j-th rule is calculated by qj(x) = Õ i=1 M m A ji (xi). Consequently.2) and m(ji. x2 . The class output of the fuzzy classi- fication system with respect to the input x can be determined by y = arg max q j ( x) × CFj j =1 R (3) 2. respectively. m( ji .2) .2) ï ç è ï è =í ï æ æ xi .1) m(ji. the member- According to the above description. PSO-Based Fuzzy Classification Systems The PSO is an evolutionary computation technique . is implemented to select an appropriate fuzzy classification system such that a low number of training patterns are misclassified by the selected fuzzy classification system. If x1 is Aj1 and x2 is Aj 2 and L and xm is Ajm . is the consequent class output of the j-th fuzzy rule. the PSO-based method is applied to select an appropriate fuzzy classification system with a low number of fuzzy rules such that the number of incorrectly classified patterns is minimized. a rule base of a fuzzy classification system can be expressed as follows [3]: j . Finally. where R is the number of fuzzy rules in the rule base. In this paper.(1) j = 1.2.

m(hI hi . I3h} = {2. if x > m h h . (4) where AIhhi .m h h ( I r i .rh. Compared with the GA.1) ÷ ÷ ø ö ÷ . In the PSO-based method. Like the GA.1) mh(ji.81]. ph = [rh gh]. the individual ph contains two parameter vectors: rh and gh. i ( I r i . the generated rule base has three fuzzy rules and {I1h. the PSO is initialized with a population of random solutions.N. where mjih = [mh(ji. x2 . That is. xi ) h Ir i r r r In order to evolutionarily determine the parameters of the fuzzy classification system.1) ïexp ç . gjh Î[0.35 0.3. n = 1. are gathered from the observation of the considered M-class classification problem. r I 2h .3) .3)] is the parameter vector to determine the membership function for the i-th input variable.2.mh ( I rh i .2) mh(ji.B) whose value is greater than or equal to 0.1] decides whether the j-th candidate rule rjh is added to the rule base of the generated fuzzy system or not. h h L .B}. This differs from many optimization methods requiring the derivation information or the complete knowledge of the problem structure and parameter. r r r é p1 ù é r1 ê ú ê ê p2 ú ê r 2 ê M ú êM P=ê ú=ê ê p ú êrh ê hú ê ê M ú êM ê p ú êr ê Lú ê L ë û ë g1 ù ú g2 ú M ú ú gh ú ú M ú gL ú ú û Then x = ( x1 .¼. i ( I r i .¼. i = 1. r I rhh } generates the premise part of the fuzzy rule base generated by the individual ph = [rh gh].2. is expressed in the following: the fuzzy rules from the candidate rules rh = [r1h r2h ¼rjh ¼rBh] so that the fuzzy rule base is generated. r = 1.2) . Consequently. {r I1h .th rule : If x1 is AIhh 1 and x2 is AIhh 2 and L and xm is AIhh m .5 is the number of fuzzy rules in the generated rule base.1) .3) î è ö ÷ ÷ ø ö ÷ ÷ ø 2 ö ÷ .2. xn2. where xn = (xn1. The parameter vector gh = [g1h g2h ¼ gjh ¼ gBh] is used to select ì æ æ x . The membership function associated with the fuzzy set AIh i is described as follows: h r m Ah (m(hI hi . m(hI hi . where B is a user-defined positive integer to decide the maximum number of fuzzy rules in the rule base generated by the individual ph. I2h. and swarm theory. individuals in the population share information between them.m. the PSO has memory so that the information of good solutions is retained by all individuals. P. yn).12 0.2) ï è è =í ï æ æ xi . rh .B) whose value is greater than or equal to 0. are the fuzzy sets of the generr ated r-th fuzzy rule. According to gh.¼.¼.¼.¼. xm ) belongs to Class H r with CF = CFr . Here. r I rh .6]. then the j-th candidate rule rjh is added to the rule base. L . Consequently.ç ç ç mh h ï ç è ( I r i . (5) r = 1. xnm) is the input vector of the n-th training pattern .2. if x £ m h h .72 0. respectively. assume that rh and gh are denoted as [r1 r2 r3 r4 r5 r6] and [0. It also requires only the information about the fitness values of the individuals in the population.6} so that {r2 r3 r6} determines the premise part of the generated rule base. Assume that N training patterns (xn. where rh represents the number h h of the fuzzy rules in the generated rule base. Its development was based on observations of the social behavior of animals such as bird flocking.L . the individual ph determines the premise part of the generated fuzzy classification system. rjh = [mj1h mj2h ¼ mjih ¼ mjMh] is the parameter vector to determine the membership functions of the j-th fuzzy rule. The individual is used to partition the input space so that the rule number and the premise part of the generated fuzzy classification system are determined. the index j of gjh (j = 1.L .1) ÷ ÷ ø (6) 2 Consequently.1) ïexp ç .5 is defined as IrhÎ {1.¼. Furthermore. the rule base of the generated fuzzy classification system is described as follows: r .ç i h ï ç ç m h ç ( I r i . In order to generate the rule base.5. For example. 2. the total number of gjh (j = 1. fish schooling. The parameter vector rh = [r1h r2h ¼ rjh¼ rBh] consists of the premise parameters of the candidate fuzzy rules.29 0. A set of L individuals.2. called population. Subsequently. each individual is represented to determine a fuzzy classification system. it has constructive cooperation between individuals.82 0.2. the consequent parameters of the corresponding fuzzy system are obtained by the premise fuzzy sets of the generated fuzzy classification system.Design of PSO-based Fuzzy Classification Systems 65 proposed by Kennedy and Eberhart [2. If gjh ³ 0.

h ÷ . t = 1. the selected fuzzy system has a low number of rules and a low number of incorrectly classified patterns simultaneously.M}. (a) Set the number of individuals (L). the maximum number of rules (B).2. and se and sr are user-defined constants for the fitness function. The procedure is described as follows: Step 1. Calculate qt. h = 1.3)¼mh(Bm. è sr ø (13) (c) Generate randomly initial velocity vectors nh.k) is defined as [m(mink ) .3) ¼ mh(B1. 2. M . t = 1. (b) Generate randomly initial population P.2. a procedure is described as follows [3]: Step 1. (7) Step 2.M} is the corresponding class output.2. M qt (10) Here.k). ji .¼. Each individual of the population is expressed as follows: ph = [rh gh].¼. Subsequently.q åq t =1 M (9) t where q= t =1 t ¹ Hr å M . Initialize the PSO-based method. (16) and æ r ö g 2 ( p h ) = exp ç . t =1 M (8) Step 3. That is. c1. Determine the grade of certainty CFr of the r-th fuzzy rule by CFr = q Hr . is randomly generated as follows: gjh = rand().2. a PSO-based method is proposed to find an appropriate individual so that the corresponding fuzzy classification system has the desired performance.2) mh(11.2.1) mh(B1. the constants for the fitness function (se and sr) and the constants for the PSO algorithm (y. In this way. i Î {1. the fuzzy classification system corresponding to the individual will satisfy the desired objective as well as possible. d1.¼. each individual corresponds to a fuzzy classification system.3}. j Î {1. In order to construct a fuzzy classification system which has an appropriate number of fuzzy rules and minimize incorrectly classified patterns simultaneously. is randomly generated as follows: m(hji .¼. ji . m(maxk ) ] and rand() is a uniformly disji .3)¼ mh(1m.2.1]. the consequent parameters of the generated fuzzy classification system are determined by the above procedure. According to the above description. k Î {1.¼. mh(ji.1) mh(11. ji . rh is the number of fuzzy rules in the rule base of the generated fuzzy classification system.1.L . d2.2) mh(Bm. j Î {1. Each velocity vector is expressed as follows: . Consequently.1) mh(1m.1].¼. the fitness function is designed to deal with the tradeoff between the number of incorrectly classified patterns and the number of fuzzy rules. (14) Thus. the fitness function is defined as follows: f h = fit ( p h ) = g1 ( p h ) g 2 ( p h ) where rh = [mh(11.3)] and gh = [g1 g2 ¼ gB]. as the fitness function value increases as much as possible based on the guidance of the proposed fitness function. the number of generations (K). and c2). ji . In order to determine the consequent parameters Hr and CFr of the r-th fuzzy rule.M for the r-th fuzzy rule as follows: qt = x p ÎClass t å qr ( x p ). NICP(ph) is the number of incorrectly classified patterns.66 Chia-Chong Chen and yn Î {1.k ) = m(mink ) + (m(maxk ) .L. tributed random numbers in [0.B}. Determine Hr for the r-th fuzzy rule by H r = arg max q t .2) mh(1m.2.1) mh(Bm. gjhÎ [0.m(mink ) ) × rand ( ).2) mh(B1. (11) (15) where fh is the fitness value of the individual ph.B}. g1(ph) and g2(ph) are defined respectively as follows: æ NICP( p h ) ö g1 ( p h ) = exp ç ÷ se è ø (12) where the range of the parameter mh(ji.

a synthetic data set and the Iris data set are employed to examine the learning ability and the generalization ability of the proposed PSO-based fuzzy classification system..5). shapes and orientations v h = v h + c1 × rand () × ( p ' best .L.2. the parameter vector ph is updated by the PSO so that the fuzzy classification system generated by the individual ph has an appropriate number of rules and a low number of incorrectly classified patterns simultaneously.3)¼ah(1m. The adjustment by Step 5 is conceptually similar to the mutation operator utilized by the genetic algorithm. Calculate the fitness value of each individual and set initial ph’.B} is randomly generated as follows: bh = j rand () .5 to the nearest integer. (a)Update p¢h. ah(ji. y = y * d2 . Example 1. fbest in the following: If f h' > f best. 4.1) ah(B1. then f best = f h' . h = 1. h =1 Set f best = fJ. (a) fh = fit (ph). round (B rand () + 0. A synthetic data set In this example.L}. gen = gen+1.Design of PSO-based Fuzzy Classification Systems 67 nh = [ah bh]. 2. (c) Set gen = 1.¼. the best solution among all the individuals obtained so far in the population is kept track of as the vector pbest associated with the global best fitness f best.2.2. Furthermore.2.L .2. and pbest. Consequently. Based on the individual pbest = [rbest gbest] with the best fitness f best. where j* = round (B rand () + 0.ph ) + c2 × rand () × ( p h .¼. (a) Update the velocity vectors in the following: Step 7. k ) Step 6.2.2) ah(11. pbeat = pJ. if gen > K then go to Step 8. pbest = ph¢. (22) (23) -m 20 min ( ji . The adjustment toward pNh and pbest by Step 5 is conceptually similar to the crossover operator utilized by the genetic algorithm. each parameter vector ph is assigned with a randomized velocity vector nh according to its own and its companions’ flying experiences so that the vector ph searches around its best vector p¢h and the global best vector pbest. p¢h = ph. (18) bjh. (21) where ah = [ah(11. Update p¢h. Set fh¢ = fh. 20 (19) Step 2. j Î{1.2) ah(B1.. . Step 4. (b)Update pbest.p h ). L. (17) p h = p h + v h . if fh > f h' . i Î{1. Each gh is randomly updated by Step 3 so that the fuzzy classification system generated by the individual ph has an appropriate number of fuzzy rules for the considered classification problem. h = 1.3)] and bh = [b1h b2h ¼bBh].. 2. 2.3}. j Î{1.. Simulation Results In this section. = 1. v h = v h × d1 . it has achieved so far in a vector p¢h..M}. which is associated with the best fitness f h' . Step 3.¼.1) ah(1m.2) ah(1m. the desired fuzzy classification system can be determined.2) ah(Bm. Each individual ph keeps track of its own best solution.1) ah(11.3) ¼ah(Bm. L. is randomly generated as follows: a (hji ... then f h' = fh. a synthetic data set containing of three clusters of various sizes. (b) Find the index J of the individual with the best fitL ness J = arg max f h . h Î{1.1) ah(Bm. (20) (b) Update the parameter vectors in the following: . h = 1.2. Step 8. k Î {1.5) rounds B rand () + 0.k ) = m max ( ji . h = 1.¼. fh’ for each individual and initial pbest. According to Step 5.2.L . as follows: If y ³ rand (). then ghj* = 1 . Decrease nh and y by the constants d1 Î [0.1] and d2 Î [0. Update the velocity vector vh and the parameter vector ph. f best.L . respectively.¼.k). h = 1.3) ¼ah(B1. Step 5.L.B}. L. ph¢ = ph.2.1].L. f ¢h.ghj*. respectively. f ¢h in the following: Calculate fh = fit(ph).L. h = 1. otherwise go to Step 3. fbest for the initial population.¼. Update the vector gh = [g1h g2h gjh¼ gBh].. k ) × rand ( ).

0.2) 0.2570 0.3604 Hj 1 3 3 2 CFj 0.1) 0. the number of generations: K = 50. j Î {1.20} : [0.8126 0.2¼.1.¼. the two-fold cross validation (2CV) is employed to examine the generalization ability of the proposed approach to classifying the iris data. Figure 1 shows the data set with a mixture of spherical and ellipsoidal clusters.5].1).1958 0 0.¼.3) 0. the range of mh(j1.2181 0.05. the constants for the fitness function: {se. In this example. the petal length in cm and the petal width in cm.75. the Iris data are separated into two subsets of the same size. where the initial conditions for the proposed method in Example 1 are given in the following: The number of individuals: L = 100.9343 Figure 1.2. the range of mh(j2. the range of mh(j2. That is.2). sr} = {5.d2} = {1. A synthetic data set with a mixture of spherical and ellipsoidal clusters in Example 1.05.20} : [0.0.5].05.5]. the range of mh(j2.¼.4406 m(j2.0. j Î {1.68 Chia-Chong Chen [12] is utilized to examine the learning ability of the proposed PSO-based fuzzy classification system. The total number of patterns in this data set is 579.¼. the sepal width in cm.3). One class is linearly separable from the other two.2. it is clear the proposed PSO-based method can select significant fuzzy rules to construct a fuzzy classification such that the number of incorrectly classified patterns are minimized.2.9637 0. .8314 0. the maximum number of rules: B = 20.20} : [0.209 0.2). The data set contains three classes.9153 0. each subset consists of Table 1.4141 0. The parameters of the selected fuzzy classification system are shown in Table 1.d1. The rule base of the selected fuzzy classification system has four rules such that the number of incorrectly classified patterns is zero. Example 2.6762 0. Simulation results of the proposed PSO-based method for the synthetic data set in Example 1. j Î {1.5].1).7361 0.¼. Following the proposed method.2) 0.0.3) 0.3087 m(j1.0154 m(j1.1) 0. each of 50 patterns.2883 0.20} : [0.5187 0. c1. simulation results of the proposed approach to classifying this data set is shown in Figure 2.455 0.2625 0. In the 2CV procedure. From simulation results of this example. j Î {1. Iris data set The Iris data set [1] contains 150 patterns with four features.20} : [0.1].10} and the constants for the PSO: {y. j Î {1.1336 m(j2.05 0. that belong to three classes (Iris Setosa.1. the latter are not linearly separable from each other.366 0.05.1]. The four features are the sepal length in cm.3121 0. j Î {1.5 0.c2.0.20} : [0. Figure 2. Iris Versicolour and Iris Virginica).3009 m(j2. 0.4669 0.2. Parameters of the selected fuzzy classification system by the proposed PSO-based method in Example 1 j 1 2 3 4 m(j1. each class refers to a type of iris plant.3).2. the range of mh(j1. the range of mh(j1.75}.

3).4].1.4.¼. Then.75}. In the proposed approach. j Î {1. 20} : [0.c2. The other subset is used as test patterns to evaluate the generated fuzzy classification system.2.6. a fitness function is defined to guide the searching procedure to select a fuzzy classification system with the desired performance.2).¼.2].¼. j Î {1. pp. The initial conditions for the proposed method in Example 2 are given in the following: The number of individuals: L = 100.¼. the obtained premise fuzzy sets are used to determine the consequent parameters of the corresponding fuzzy classification system.” Proc.67 % 96.edu/~mlearn/ [2] Eberhart. and Merz.1.87% Average classification rate on test patterns 96. j Î {1. UCI Repository of Machine Learning Database.1. j Î {1. The simulation results show that the selected fuzzy classification system not only has an appropriate number of rules for the considered classification problem but also has a low number of incorrectly classified patterns.2. j Î {1. the range of mh(j2.7]. Consequently. http://www. The above procedure is iterated 20 times using different partitions of the 150 patterns into two subsets for calculating the average classification rates on these data.1.. Keogh. 20} : [0.1.¼. each individual ph consists of two parameter vectors: rh and gh The parameter vector gh is updated so that the generated fuzzy classification system has an appropriate number of fuzzy rules.75]. j Î {1.10 4. Subsequently.1).. Nagoya Japan.1).uci.2. the range of mh(j1. j Î {1.2). Oct. the range of mh(j3.75. E.65]. Micro Machine and Human Science.725 tion system has high generalization ability for the classification problem of the Iris data set.¼. the range of mh(j1. The same training-and-testing procedure is also followed after the roles of the subsets are exchanged with each other. Simulation results of the proposed PSO-based method for the Iris data set using 2CV method in Example 2 Average classification rate on training patterns 98..3).20} : [0. R.0.¼. j Î {1.4. California.00 597. The parameter vector rh is updated so that the premise part of the generated fuzzy classification system has appropriate membership functions for the considered classification problem. Generalization ability for test patterns in the Iris data classification using the same 2CV method Method Pruning Multi-rule-table GA-based Proposed method Average classification Average number rate on test patterns of fuzzy rules 93.20} : [2.20} : [1.2].20} : [0. “A New Optimizer Using Particle Swarm Theory.75 10.1.¼. 39-43 (1995). Sym.80 % 28.20} : [0.2. the range of mh(j4. the range of mh(j3.8% Average number of fuzzy rules 4. j Î {1.1). 5. 75 patterns.1.2).1. C. j Î {1.75]. it is obvious that the proposed PSO-based fuzzy classifica- Acknowledgement This research was supported in part by the National Science Council of the Republic of China under contract NSC 93-2218-274-001 References [1] Blake.20} : [0.¼.2.10} and the constants for the PSO: {y.1.20} : [0.1. j Î {1.2.2. sr} = {5. the range of mh(j3.20} : [0.2. the range of mh(j4.65].1.3). C. From simulation results for the Iris data set. The average number of fuzzy rules of the selected fuzzy classification systems and the average classification rates on training patterns and test patterns are summarized in Table 2.1.7]. the range of mh(j1. the range of mh(j2.2.5].725 Table 3. j Î {1..d2} = {1.1.d1.3). Conclusions A PSO-based method is proposed to design an appropriate fuzzy classification system for pattern classification.2. the range of mh(j4.c1.2.¼.20} : [0.2].¼.1.2). the constants for the fitness function: {se.2.1.2].7.30 % 94.20} : [4.2.¼. Irvine (1998). Table 3 compares our results with the results in [7] for the classification of the Iris data using the same 2CV method. .1). Int. and Kennedy. each individual corresponds to a fuzzy classification system.ics.2. the number of generations: K = 50. One subset is used as training patterns to construct a fuzzy classification system by the proposed approach. J. J.2. the range of mh(j2. the maximum number of rules: B = 100.2.Design of PSO-based Fuzzy Classification Systems 69 Table 2.0.1. Univ.30 % 90.1.

and Tanaka. pp. C. Sun.” IEEE Trans. J. Vol. K. C. Neural Networks. E.” IEEE Trans. C. 776-786 (1992). H. X. C. J.” IEEE Trans. P. 238-250 (1996).. C. 1942-1948 (1995). “A Hybrid Clustering and Gradient Descent Approach for Fuzzy Modeling. “A Novel Algorithm for Data Clustering. C. and Su. 22. [7] Nozaki.. 425-442 (2001). Neural Networks. Conf. on Systems. pp. Prentice Hall.S. R.. [10] Wong. pp. Vol. (1997)... Man and Cybernetics. “Fuzzy Min-Max Neural NetworksPart 1: Classification.A. Vol. and Eberhart. [12] Wong. M. Vol. “Generating Fuzzy Rules by Learning from Examples.. 34.S. 686-693 (1999). on Systems.. Essentials of Fuzzy Modeling and Control. Ishibuchi. “Adaptive Fuzzy Rule-Based Classification Systems. Perth. 904-911 (2000). Vol. Vol.. pp. S. pp. 3. “ANFIS: Adaptive-Network-Based Fuzzy Inference systems. on Fuzzy Systems. Man and Cybernetics-Part B: Cybernetics. on Systems. and Tanaka. Neuro-Fuzzy and Soft Computing. 3. R. [9] Wang.. D. Sep.” IEEE Trans. pp. “Selecting Fuzzy If-Then Rules for Classification Problems Using Genetic Algorithms. and Mendel. H. C. 14141427 (1992).” IEEE Trans. H. 2005 . [6] Kennedy.” IEEE Trans. [5] Jang.. J.. “A GA-Based Method for Constructing Fuzzy Systems Directly from Numerical Data. [13] Yager.” Pattern Recognition. K. and Chen. Vol. Man and Cybernetics-Part B: Cybernetics..70 Chia-Chong Chen [3] Ishibuchi.” IEEE Trans.A. pp. [4] Jang. Manuscript Received: Apr. C. Australia. [11] Wong. IEEE Int.. 23. Fuzzy Systems. on Systems. Nozaki. New Jersey. L. C... C. Man and Cybernetics. Chen. U. 27. and Chen. 260-270 (1995).” Proc. N. K. S. 13. 2005 Accepted: Jul.. 30. C. 3. Vol. J. 29. Yamamoto. Aug. U. and Filev. C. (1994). pp. and Mizutani. No. P. T. 665-685 (1993). pp. H. C. “Particle Swarm Optimization. [8] Simpson. M. R. 4. C. John Wiley. New York..