Abstract: A new +L-R feature selection algorithm is proposed which combines trace ratio criterion selection and +L-R method based on error region, it uses trace ratio selection to obtain a optimal subset and uses error region by +L-R selection to get a new feature which can classify error sample efficiently in the region of error samples and error classification rate can be decreased efficiently. Using +L-R algorithm can reduce data redundancy. Experimental results show that the proposed algorithm improves trace ratio criterion significantly and lower error rate can be achieved at the same time.

Key words: error region, trace ratio criterion, feature selection, machine learning, pattern recognition

摘要： 提出一种将迹比准则和基于错分区域的+L-R方法相结合的特征选择算法。该算法使用迹比算法得到优秀特征子集，对分类产生的错分区域进行+L-R选择得到新特征，新特征可以区分之前被错分的数据，从而降低错分率。采用+L-R算法降低数据冗余。实验结果表明，该算法有效改进迹比准则特征选择算法，同时降低错分率。

关键词: 错分区域, 迹比准则, 特征选择, 机器学习, 模式识别

CLC Number: 

• TP18