Metalearning for choosing feature selection algorithms in data mining: Proposal of a new framework
作者：
Highlights：
• A new Metalearning architecture proposal to recommend Feature Selection algorithms.
• FS algorithms are ranked using a new multicriteria performance measure.
• The proposed architecture has low computational cost and is human understandable.
• Evaluation on 150 data sets from literature review and four well-known FS algorithms.
• Accuracy higher than 90% was obtained in the recommendation of FS algorithms.
摘要
•A new Metalearning architecture proposal to recommend Feature Selection algorithms.•FS algorithms are ranked using a new multicriteria performance measure.•The proposed architecture has low computational cost and is human understandable.•Evaluation on 150 data sets from literature review and four well-known FS algorithms.•Accuracy higher than 90% was obtained in the recommendation of FS algorithms.
论文关键词：Data quality,Algorithm recommendation,Symbolic metamodels,Machine learning
论文评审过程：Received 1 October 2015, Revised 4 November 2016, Accepted 15 January 2017, Available online 16 January 2017, Version of Record 25 January 2017.
论文官网地址：https://doi.org/10.1016/j.eswa.2017.01.013
