Feature selection based on feature interactions with application to text categorization
作者：
Highlights：
• A higher-order interaction based feature selection method FJMI is proposed.
• FJMI employs five-dimensional joint mutual information to capture interactions.
• FJMI is applied to text categorization and improves the highest accuracy by 9%.
• FJMI is evaluated on eleven competing methods and twenty-five data sets.
• FJMI performs better or equally well than other method in about 80% of the cases.
摘要
•A higher-order interaction based feature selection method FJMI is proposed.•FJMI employs five-dimensional joint mutual information to capture interactions.•FJMI is applied to text categorization and improves the highest accuracy by 9%.•FJMI is evaluated on eleven competing methods and twenty-five data sets.•FJMI performs better or equally well than other method in about 80% of the cases.
论文关键词：Feature selection,Feature interaction,Mutual information,Joint mutual information,Text categorization
论文评审过程：Received 2 January 2018, Revised 9 November 2018, Accepted 10 November 2018, Available online 10 November 2018, Version of Record 27 November 2018.
论文官网地址：https://doi.org/10.1016/j.eswa.2018.11.018
