Abstract
In recent years, the pattern recognition is used in various fields. In the machine learning for the pattern recognition, it is necessary to consider an appropriate configuration of various processes such as the feature extraction from data and the feature selection. In the selection of extracted features, a versatile method to find a combination of features effective for the identification has not been developed. The multiple feature selection method which automatically selects features in response to the input data by using various classifiers with different features has been proposed. In the previous research, ID3 ensemble system that performs the multiple feature selection by constructing a decision tree has been proposed. The previous research demonstrated that this method is effective to enhance the efficiency of the computation for the multidimensional input. However, constructed decision trees were more likely to have a complicated structure such as the excessive classification rules for specific data. This tendency has probabilities to reduce the generalization performance for unknown data. Therefore, this study attempts to investigate an effective method of multiple feature selection for the multidimensional input. Several numerical examples are presented to demonstrate the effectiveness of multiple feature selection method based on the decision tree.
