集成学习的多分类器动态融合方法研究

AdaBoost集成学习方法中,分类器一经学习成功,其投票权值就已确定,同一分类器对所有待测样本均有相同的投票权值。对于难于分类样本,具有良好分类性能的少数分类器权值却较低。提出适用于集成学习方法的权重自适应调整多分类器集成算法。根据多分类器行为信息,产生待测样本局部分类精度的有效判定区域,基于有效判定区域选择不同的分类器组合,并调整其相应权重,利用样本集上的统计信息来动态指导分类集成判决。实验结果表明,该算法提高了集成分类性能。

Study of integration method for multiple classifiers on ensemble learning

As soon as a classifier is trained by AdaBoost ensemble learning algorithm,it has a constant weight for all test instances.A few of classifiers which have better classification performance for some instances hard to classified have usually small weights.A new dynamic weight self-adjusting algorithm is presented for ensemble learning method.The effective determining area of the test instance is computed automatically based on the classification behavior of classifiers.Some combine classifiers are selected and their weights are adjusted based on the effective determining area of the test instance.An integration decision is made by using of the statistics information of sets of instances.The experiment result shows that ensemble classification performance is improved by use of this algorithm.