Abstract: In order to eliminate adverse effect brought by bad classifiers and improve effect and stability of combiners，an approach extracting classifiers based on Information Gain（IG） is proposed.It can reduce classifier space with high dimension，and then learn a combiner in lower dimension.The compared results obtain on multiple public avaiable datasets show that the method is feasible firstly.Secondly，its performance is perfect.

摘要： 为了减少表现差的个体分类器对集成器分类性能的影响，提高集成器分类效果及稳定性，提出了基于信息增益的分类器选择方法。该方法将高维分类器空间压缩至低维分类器空间，并在该空间内学习集成器。在多个数据集上的比较实验结果表明，该方法可行，其集成性能较理想。