Feature selection based on network maximal correlation

Abstract

Feature selection can effectively increase the accuracy of machine learning and improve the efficiency of the algorithm. Therefore, it has emerged as a critical technology related to data mining, machine learning, and has shown great impacts in many applications, including biomedical, financial and communication. However, the increase in data dimension poses a serious challenge to many existing feature selection methods in terms of effectiveness and efficiency. Hirschfeld-Gebelein-Renyi maximal correlation is a effective measure of the correlation between variables. In this paper, with this measure, we proposed a improved Network Maximal Correlation (NMC) model. It can quickly and effectively calculate the statistical dependence between feature set and label variable. Further, based on the Recursive Feature Elimination (RFE) algorithm, a new NMC-RFE feature selection method is Further proposed. The experimental results show that the proposed method can obtain much better feature subsets from high dimensional data sets with faster calculation speed and better accuracy.