基于属性约简的PLS加权朴素贝叶斯分类

Weighted naive Bayes classifier based on attribute reduction-PLS

朴素贝叶斯算法是一种简单而高效的分类算法,它的属性独立性假设,影响了它的分类性能.针对这种问题,在分析属性相关性的基础上,通过属性约简选择一组近似独立的属性约简子集,提出一种基于属性约简的偏最小二乘回归加权朴素贝叶斯分类算法.对不同的条件属性给予不同的权值,从而在保持简单性的基础上有效地提高了朴素贝叶斯分类算法的分类性能.实验结果表明,该方法可行且有效.

Naive Bayes classifier is a simple and effective classification method, but its attribute independence as- sumption affect its classification performance. In response to this problem,based on the analysis of attribute correla- tion,a set of approximate independent subset of attribute reduction is selected by attribute reduction,weighted naive Bayes classifier based on attribute reduction-PLS is proposed. Different conditions attribute give different weights, thus it effectively improve the classification performance of naive Bayes algorithm on the basis of simplicity. Experi- mental results show that the method is feasible and effective.