朴素贝叶斯分类器建立在其数据"特征值之间相互条件独立"的基础上,而在实际应用中该假设难以完全成立.针对这种现象提出一种算法,即通过寻找对产生错误分类影响最大的特征值,并依此特征值的关联项对数据项扩充,在此基础上对扩充项添加权重,以达到提升分类器精度的效果.最后对权重的大小加以论证,实验分析了不同大小的权重对分类器正确率的影响.实验结果表明,添加关联项扩充训练集,可以有效提升贝叶斯分类器的正确率.%Naive Bayes classifier is based on the hypothesis that parameters of the sample are mutually conditional independent.The practical application of this hypothesis is hard to established, so this paper proposes a new algorithm to improve Naive Bayes classifier through looking for properties that have the maximum influence on error classification with an effectively way, finding related items to extend the original data set, then adding weights to the related items.This paper shows the results by experiments, and related items make the classifier work better.
展开▼
