基于交叉熵方法的选择性AODE算法

Learning Selective AODE for Classification Based on Cross-Entropy Method

AODE(Averaged One-Dependence Estimators)算法是最近提出的一种典型的基于naveBayes的改进算法,并受到国际机器学习界的关注。交叉熵方法(Cross-entropy Method)是一种解决组合优化问题的全局随机搜索算法,已经成功地被应用到许多经典的NP问题中。给出了AODE算法选择性集成的理论基础,并基于交叉熵方法,提出了解决AODE算法选择性集成的CESAODE(Cross-Entropy method for Selective AODE)算法。在WEKA平台上使用UCI数据集进行的仿真实验结果表明,CESAODE算法比现有的分类算法,例如AODE等具有更好的分类性能。

AODE （Averaged One-Dependence Estimators） is a recently proposed semi-naive Bayes algorithm and has attracted the attention of the machine learning community. Cross-Entropy Method is a recently proposed random search algorithm and has been successfully applied into a wide range of NP hard problem with promising result. The selective AODE problem was studied and the theoretical foundation was given to explain why model selection for AODE was useful, and CESAODE （Cross-Entropy method for Selective AODE） was proposed which could efficiently search the optimal subset over the whole one-dependence estimators of AODE. The experimental results show that the algorithm significantly outperforms the existing algorithms such as AODE in term of classification accuracy.