Designing simulated annealing and subtractive clustering based fuzzy classifier

A new systematic way in order to obtain optimized fuzzy inference system for classification task is developed. The proposed algorithm, Simulated Annealing (SA) and Subtractive Clustering (SC) based Fuzzy Classifier (SASCFC) is a cooperation of the SA and the SC methods. The SA is used in order to optimize the SC parameters, feature subspace and output threshold value of fuzzy based classifier. A hybrid feature selection strategy which combines filter and wrapper type approaches is also proposed. In order to demonstrate the effects of these optimizations, we proposed four different SASCFC classifiers which are called as the SASCFC-Type1, Type2, Type3 and Type4. The performance and rule base complexity of proposed classifiers are compared with each other and also some classifier tools on some well known classification tasks. The results show that the proposed classifiers have a satisfactory performance in comparison with its counterparts.