论文部分内容阅读
为了提高传统AdaBoost(adaptive boosting)算法的收敛性能,提出一种基于多步校正的AdaBoost改进算法。在该算法中,训练样本的分布更新不仅与当前分类器有关,而且也需要考虑到前面的若干分类器;进一步地,新的算法在每一个分类器集成进来后会对前面产生的某些分类器权重进行修正。在UCI数据集Diabets,Heart-statlog和Breast cancer Wisconsin上的实验表明,该算法获得了更好的训练误差和测试误差的优化性能。这说明,利用多步校正策略不但可以提高成员分类器的搜索效率,而且可以进一步地改进集成分类器的整体性能。
In order to improve the convergence performance of traditional AdaBoost algorithm, an improved AdaBoost algorithm based on multi-step correction is proposed. In this algorithm, the updated distribution of training samples is not only related to the current classifier, but also needs to consider several previous classifiers; furthermore, the new algorithm integrates each classifier into some classifiers Weight correction. Experiments on UCI datasets Diabets, Heart-statlog and Breast cancer Wisconsin show that this algorithm achieves better performance of training error and test error optimization. This shows that the use of multi-step correction strategy can not only improve the search efficiency of the member classifier, but also can further improve the overall performance of the integrated classifier.