一种新的互信息特征子集评价函数

摘要

传统基于互信息的特征选择方法较少考虑特征之间的关联,并且随着特征数的增加,算法复杂度过大,基于此提出了一种新的基于互信息的特征子集评价函数.该方法充分考虑了特征间如何进行协作,选择了较优的特征子集,改善了分类准确度并且计算负荷有限.实验结果表明,该方法与传统的MIFS方法相比较,分类准确度提高了3％～5％,误差减少率也有25％ - 30％的改善.%Conventional mutual-information-bascd feature selection algorithms seldom considers how features work together, with the features incresement, the computational complexity of the algorithms will increase dramatically.So propose mutual-information-based evaluation function for feature subset selected, different from other mutual-information-based feature selection algorithm,it considers how features work together.So it produces the optimal feature subset and improves the classification accuracy of classifier.The computational complexity of it is also limited.The results show about 3% ~ 5% increase for classification accuracy and 25% ~ 30% improvement for error reduction rate compare with other conventional mutual-information-based feature selection algorithms.