ATTRIBUTE REDUCTION OF NUMERICAL DECISION TABLE BASED ON FUZZY DISTANCE USING HEURISTIC ALGORITHM

Abstract

Attribute reduction of the decision table is the process of selecting a subset of the conditional attribute set that preserves the degree of information measurement. Attribute reduction of numerical decision table according to fuzzy rough set approach has studied extensively in recent years. Attribute reduction methods according to the fuzzy rough set approach inherits traditional rough set methods with fuzzy equivalent relations replacing the equivalent relations. The value of fuzzy similarity is in the range [0, 1] which shows the close or similar properties of two objects. The fuzzy equivalent relations process directly on numeric value domain without through steps of discrete data. The original method is a fuzzy positive region. Researchers have developed methods using fuzzy entropy and fuzzy distance to improve the quality of the classification accuracy and reduce the execution time of the algorithm. This paper proposed a fuzzy distance and constructed a heuristic algorithm to find one reduction set of numerical decision table which called the reduct. The proposed method preserves information measurement of the conditional attribute set. Experiments on datasets taken from the UCI repository show that the proposed method improves the quality of classification accuracy and execution time of the algorithm compared with the methods using fuzzy positive region and fuzzy entropy on most experimental datasets.