Feature selection by genetic algorithm that adopts the principle of margin-maximization and its application to character recognition

This paper proposes a feature selection method that improves the recognition rate significantly for not only training samples but also unknown samples by using the principle of margin-maximization in the support vector machine (SVM). SVM is well-known as a recognition method that can discriminate unknown samples with high precision, so feature selection with high recognition rates for unknown samples can be expected by adopting the principle of margin-maximization, the technical basis of SVM. We perform experiments on five sets of hand-written Kanji character patterns. Each set consists of patterns in two similar categories. The results show that the proposed method improves the recognition rate significantly for not only training samples but also the unknown samples as expected.