A hybrid feature selection approach by combining miD and miQ


Feature selection is one of the most important steps in pattern recognition, data mining, machine learning and computer vision. The main purpose of feature selection is dimensionally reduction, improve the performance of the classifier, improve learning efficiency and enable better understanding of the underlying process. In this paper we proposed a hybrid technique by combining Mutual Information Difference (miD) and Mutual Information Quotient (miQ) for feature selection. We tested our technique on two continues data sets of gene expression profiles; the Leukemia and the Colon cancer; using three classifiers, which are SVM (Support Vector Machine), Naive Bayes (NB) and Neural Network (NN). We compared our proposed technique with miD and miQ and it outperforms others. The SVM perform best among three classifiers.

Proceeding of the ITFE Summer Conference, South Korea, 1:367-373.