摘要
本文讨论了分类器组合的叠加法以及相应的叠加策略,提出了一种新的基于叠加法的元学习策略。该策略对基分类器的预测结果进行投票表决,将表决的结果作为1层训练数据。实验结果表明,该方法比简单平均化验概率法的分类效果要好。
This paper describes stacking method and relative strategies, and proposes a new meta-learning strategy based on stacking framework. It votes the predictive results of basic classifier, and then takes the voting results as training data of level-1. Experiment results show that this approach can obtain more definite information from classifier outputs and achieves better classification effect than the weighted approach based on averaging posterior probability.
出处
《通讯和计算机(中英文版)》
2006年第4期26-29,共4页
Journal of Communication and Computer
基金
本文得到广西“新世纪十百千人才工程”专项(桂人字2001213)基金项目和广西自然科学基金项目(桂科字0229008)的资助.