Feature subset evaluation using fuzzy measures

Feature selection is an important prerequisite of any pattern recognition system. For the selection of good features, one has to use some criterion for the assessment of its quality. Generally, a subset of M features are needed to be selected from all possible combinations of M features out of N features. In this paper, a measure for the evaluation of the effectiveness of a feature subset has been proposed with the help of fuzzy measures as an alternative to statistical measures. This measure, in conjunction with the branch-and-bound technique, can be used to find out the best possible feature subset from all possible subsets. The algorithm has been implemented on different data sets to explain its capability. The proposed measure is computationally easy and is suitable for use in a near-optimal search technique