Class-specific mutual information variation for feature selection
作者：
Highlights：
• A novel feature selection method is proposed based on information theory.
• A new term calculating dynamic information of selected features is proposed.
• We redefine the feature relevancy.
• We implement experiments on 20 benchmark data sets.
• Our method outperforms seven competing methods in terms of accuracy.
摘要
•A novel feature selection method is proposed based on information theory.•A new term calculating dynamic information of selected features is proposed.•We redefine the feature relevancy.•We implement experiments on 20 benchmark data sets.•Our method outperforms seven competing methods in terms of accuracy.
论文关键词：Feature selection,Information theory,Dynamic change,Classification
论文评审过程：Received 28 June 2017, Revised 5 February 2018, Accepted 18 February 2018, Available online 19 February 2018, Version of Record 26 February 2018.
论文官网地址：https://doi.org/10.1016/j.patcog.2018.02.020
