A KNN-BASED PATTERN SELECTION METHOD

摘要：

This paper presents a new pattern selection method to select critical patterns from training data set. The kNN-based pattern selection method select a new type of critical patterns, which are termed periphery patterns, besides border and edge patterns. It is applicable to train those types of classifiers which require spatial information of the training data set. The proposed method, which is named BEPP algorithm, is evaluated on benchmark problems using support vector machines, and nearest neighbours classifiers. The experiment results show that the selected patterns are able to represent the data distribution boundary in input space.

关键词： Pattern selection data condensing edge patterns border patterns periphery patterns

作者: Y.X. He Y.L. Yu

作者单位: Tongji University, Shanghai, China

会议类型: 国际会议

会议名称: 2012 International Conference on System Simulation(2012年国际系统仿真学术会议)

会议地点: 上海

会议语种:英文

页码: 520-523

在线出版日期: 2012-04-06（万方平台首次上网日期，不代表论文的发表时间）