基于最小最大模块化集成特征选择的改进

摘要

With the expansion of the data size,a single weak classifier has been unable to predict unknown samples accurately. To solve this problem,an integrated learning is proposed. Combined the integrated learning and classification,the idea of integration is also used in the feature selection at the same time. For the increase of sample prediction accuracy,a strategy based on Min-Max-Module (M3) is put forward. It makes integrated learning applied to feature selection algorithms and classifier,and compares four kinds of integration strategies as well as three different classification methods. The results show that the proposed method can be able to achieve good results in most ca-ses,and can well handle imbalanced data sets.%随着数据规模的扩大,单个弱分类器的准确率已经无法很好地对未知样本进行预测,为此提出了集成学习。在集成学习与分类器结合的同时,集成的思想同样被用到了特征选择中。从提高对样本预测的准确率的角度出发,提出一种基于最小最大模块化(Min-Max-Module,M3)的策略。它同时将集成学习应用到了特征选择算法和分类器中,并对比了四种集成策略以及三种不同的分类方法。结果表明,提出的方法在大多情况下能取得不错的效果,并且能很好地处理不平衡的数据集。