An EnKF-based scheme to optimize hyper-parameters and features for SVM classifier
作者：
Highlights：
• We adopt the EnKF based scheme to optimize hyper-parameters and features.
• We build the EnKF based framework for parameter optimization.
• Multiple ensemble and ensemble evolution are proposed to enhance effects.
• Householder method is proposed to optimize the EnKF analysis computation.
• The EnKF scheme outperforms three current Bayesian optimization methods.
摘要
Highlights•We adopt the EnKF based scheme to optimize hyper-parameters and features.•We build the EnKF based framework for parameter optimization.•Multiple ensemble and ensemble evolution are proposed to enhance effects.•Householder method is proposed to optimize the EnKF analysis computation.•The EnKF scheme outperforms three current Bayesian optimization methods.
论文关键词：EnKF,SVM,Hyper-parameter optimization,Feature weighting,Feature selection
论文评审过程：Received 23 December 2014, Revised 28 June 2016, Accepted 15 August 2016, Available online 17 August 2016, Version of Record 24 September 2016.
论文官网地址：https://doi.org/10.1016/j.patcog.2016.08.014
