Dynamically Threshold Value Determination in the Optimal Fuzzy-Valued Feature Subset Selection

Abstract

Feature subset selection is a pattern recognition problem which is usually viewed as a data mining enhancement technique. By viewing the imprecise feature values as fuzzy sets, the information it contains would not be lost compared with the traditional methods. Optimal fuzzy-valued feature subset selection (OFFSS) is a technique for fuzzy-valued feature subset selection. The core of OFFSS is the heuristic search algorithm for finding a path in the extension matrix where elements are the overlapping degree of two fuzzy sets. The path is all the elements less than or equal to a certain threshold value. Different threshold values would seriously affect the quality of the feature subset. The method of determining the threshold value has not been discussed in OFFSS. This paper focuses on the problem of determining the threshold value dynamically in OFFSS. By applications of the result feature subset to fuzzy decision tree induction and by comparison with the original algorithm, the revised algorithm is demonstrated more satisfying training and testing accuracy in the selected five UCI standard datasets.