Variable selection based on maximum information coefficient for data modeling
Fuchang Chu, Zhenping Fan, Baohui Guo, Dan Zhi, Zijian Yin, Wenjie Zhao · 2017
Whether the variable selection is accurate or not affect the accuracy and generalization ability of the model. The traditional variable selection method is difficult to maintain a high stability under high collinearity. In order to solve the problem, we propose a new method MICFS (Feature Select based on Maximal Information Coefficient), which combines the maximum information coefficient with the existing mutual information variable selection method. Firstly, this paper introduces the theory of mutual information and the variable selection algorithm based on mutual information, and then use the maximum information coefficient instead of the original mutual information criterion. Finally, the validity of method is verified by using the Friedman data set. The result shows that this method can meet the requirements of variable selection in a high collinearity and high noise environment.