Statistical computation of feature weighting schemes through data estimation for nearest neighbor classifiers
作者：
Highlights：
• Redundant features may deteriorate the performance of the Nearest Neighbor rule.
• This paper proposes a new feature weighting classifier to overcome this problem.
• Imputation methods are used to estimate a distribution of values for each feature.
• A statistical test sets weights based on differences of the feature distributions.
• Our proposal outperforms the rest of the classifiers considered in the comparisons.
摘要
Highlights•Redundant features may deteriorate the performance of the Nearest Neighbor rule.•This paper proposes a new feature weighting classifier to overcome this problem.•Imputation methods are used to estimate a distribution of values for each feature.•A statistical test sets weights based on differences of the feature distributions.•Our proposal outperforms the rest of the classifiers considered in the comparisons.
论文关键词：Feature weighting,Imputation methods,Nearest neighbor,Classification
论文评审过程：Received 21 March 2013, Revised 14 May 2014, Accepted 14 June 2014, Available online 23 June 2014.
论文官网地址：https://doi.org/10.1016/j.patcog.2014.06.012
