Multiple Feature Selectoin for Pattern Recognition Using ID3 Ensemble System

In pattern recognition, feature selection is a quite important process for constructing practical systems. However, because there are many features as candidates to recognize object, it is difficult to select appropriate features for pattern recognition systematically. the previous research proposed a pattern recognition method using the ensemble system based on fuzzy classifier for multiple feature selection. However, this method can not apply for the problems with many input vectors because it takes a lot of time for learning when the number of the input vector increases. in this study, an attempt is made to overcome the problem by introducing ID3 (Iterative Dichotomizer 3) with classifiers consisting of many feature vectors. ID3 constructs a decision tree for multiple feature selection with the results obtained from classifiers based on each feature. Therefore, it is possible to select appropriate features applied many input vectors. Several benchmark problems are presented to demonstrate the efficiency and applicability of the proposed method.