Construction of fuzzy knowledge bases incorporating feature selection

Constructing concise fuzzy rule bases from databases containing many features present an important yet challenging goal in the current researches of fuzzy rule-based systems. Utilization of all available attributes is not realistic due to the "curse of dimensionality" with respect to the rule number as well as the overwhelming computational costs. This paper proposes a general framework to treat this issue, which is composed of feature selection as the first stage and fuzzy modeling as the second stage. Feature selection serves to identify significant attributes to be employed as inputs of the fuzzy system. The choice of key features for inclusion is equivalent to the problem of searching for hypotheses that can be numerically assessed by means of case-based reasoning. In fuzzy modeling, the genetic algorithm is applied to explore general premise structure and optimize fuzzy set membership functions at the same time. Finally, the merits of this work have been demonstrated by the experiment results on a real data set.