Hybrid Multistage Feature Selection Method and its Application in Chinese Medicine

Abstract

The experimental data on traditional Chinese medicine efficacy has many irrelevant and redundant features, and different feature combinations have different effects. Therefore, we propose a hybrid multistage feature selection algorithm based on approximate Markov blanket and improved black widow algorithm. The first stage remove irrelevant features by the maximum information coefficient. The second stage delete redundant features from clustered searched by approximate Markov blanket by Lasso algorithm to avoid information loss. The third stage search the optimal feature subset by improved black widow algorithm that used the fast reproduction strategy, the child eating mother strategy and the population restriction strategy. The proposed approach is tested on the basic material data of traditional Chinese medicine and 9 UCI datasets, and compared with other feature selection algorithms. The experimental results show that the algorithm can obtain a small number of feature subsets with high accuracy, and has good time performance.