Cost-sensitive classifier chains: Selecting low-cost features in multi-label classification
作者：
Highlights：
• Novel method which incorporates the feature cost information into the learning process is proposed.
• The proposed method (CSCC) combines classifier chains and penalized logistic regression with a modified elastic-net penalty which takes into account costs of the features.
• We also propose an adaptive version (A-CSCC) in which penalty factors are changing during fitting the consecutive models in the chain.
• We prove the stability and provide a bound on generalization error of our algorithm.
• The method is successfully applied on real medical datasets for which cost information is provided by experts.
• We propose an experimental framework in which features are observed with measurement errors and the costs depend on the quality of the features.
摘要
•Novel method which incorporates the feature cost information into the learning process is proposed.•The proposed method (CSCC) combines classifier chains and penalized logistic regression with a modified elastic-net penalty which takes into account costs of the features.•We also propose an adaptive version (A-CSCC) in which penalty factors are changing during fitting the consecutive models in the chain.•We prove the stability and provide a bound on generalization error of our algorithm.•The method is successfully applied on real medical datasets for which cost information is provided by experts.•We propose an experimental framework in which features are observed with measurement errors and the costs depend on the quality of the features.
论文关键词：Multi-label classification,Cost-sensitive feature selection,Classifier chains,Logistic regression,Stability,Generalization error bounds
论文评审过程：Received 29 December 2017, Revised 25 July 2018, Accepted 27 September 2018, Available online 28 September 2018, Version of Record 4 October 2018.
论文官网地址：https://doi.org/10.1016/j.patcog.2018.09.012
