Combining localized fusion and dynamic selection for high-performance SVM

作者：

Highlights：

• We present multiple SVMs combined by localized fusion and dynamic selection.

• Clustering OVA SVMs trained for each class constructs multiple decision templates.

• A naïve Bayes classifier selects the localized models dynamically for a new sample.

• Experiments on nine benchmark datasets show its superiority against alternatives.

• It proves to effectively manage the unbiased-variance and bias in real datasets.

摘要

•We present multiple SVMs combined by localized fusion and dynamic selection.•Clustering OVA SVMs trained for each class constructs multiple decision templates.•A naïve Bayes classifier selects the localized models dynamically for a new sample.•Experiments on nine benchmark datasets show its superiority against alternatives.•It proves to effectively manage the unbiased-variance and bias in real datasets.

论文关键词：Hybrid approach,Classifier fusion,Dynamic selection,Sub-class modeling,Support vector machines

论文评审过程：Available online 25 July 2014.

论文官网地址：https://doi.org/10.1016/j.eswa.2014.07.028