关键词: 特征选取, 信息增益, 样本权重, 噪声干扰

Abstract:

According to the characteristics of gene expression data, a gene feature selection model based on improved information gain was put forward. The improved information gain was proposed to measure gene information quantity with sample weight and a no de-noising and de-noising gene feature selection model was established. The proposed model is compared with common gene selection model using four classifiers. Experiments validate that the proposed method can improve stability of feature selection algorithms without sacrificing predictive accuracy.

Key words: feature selection, information gain, sample weight, noise interference