Fuzzy systems have been successfully applied to various classification problemsdue to its capacity of dealing with uncertain and fuzzy information. In manyapplication tasks, fuzzy rules are manually derived from human expert knowledge.Since the main drawback of these approaches is that the expert knowledge is not easyto acquire, the automatic derivation of the linguistic fuzzy rules from numerical datahas been proposed to face the problem.This paper proposes a method to design fuzzy classification system based on antcolony algorithm. In order to obtain fuzzy models with a goodinterpretability-accuracy trade-off, the method sorts out a group of input variablesfrom training patterns with high performance of identification. Then the patternspaces are partitioned by a fixed fuzzy grid. After the identification of the surfacestructure, ant colony algorithm is applied to identify the parameters of the MFs usedin the rule base, taking into account the information provided by training patterns.Target function presents positive feedback after each iteration. Ant colony algorithmoptimizes the partition of input space to obtain a fuzzy model with smaller number ofinput variables and fuzzy rules.This paper presents Max-Min Ant System (MMAS), a good alternative toexisting algorithms, to optimize the fuzzy modeling. The range of possible pheromonetrails on each solution component is limited to an interval to avoid stagnation of thesearch.The performance of the proposed method both for training and test data isexamined by computer simulations on the Iris and Wine data classification problems.The obtained results lead us to remark the good performance of the proposal in termsof interpretability, accuracy, and efficiency. |