Efficient feature size reduction via predictive forward selection
作者：
Highlights：
• A new approach for faster feature selection using experience is presented.
• Comparison to wrapper, filter and embedded approaches on 59 datasets is provided.
• The difference to forward selection is statistically not significant.
• It requires much fewer classifier evaluations.
• It is statistically significant better than filter and embedded approaches.
摘要
Highlights•A new approach for faster feature selection using experience is presented.•Comparison to wrapper, filter and embedded approaches on 59 datasets is provided.•The difference to forward selection is statistically not significant.•It requires much fewer classifier evaluations.•It is statistically significant better than filter and embedded approaches.
论文关键词：Feature selection,Meta-learning,SVM,Classification,Forward selection
论文评审过程：Received 12 November 2012, Revised 3 June 2013, Accepted 8 October 2013, Available online 18 October 2013.
论文官网地址：https://doi.org/10.1016/j.patcog.2013.10.009
