基于属性约简的加权朴素贝叶斯分类算法

Weighted Naive Bayes Classification Algorithm Based on Attribute Reduction

朴素贝叶斯在处理分类问题上简单高效,通常它假设属性间是条件独立的,且各属性变量对类变量的影响程度是相同的,但在实际应用中这些都难以被满足,从而使得其分类性能降低.因此,提出基于属性约简的加权朴素贝叶斯分类算法,该算法首先根据各属性不同取值的分类能力及属性间的对称不确定性大小,去除了无关属性和冗余属性,使得筛选后的属性之间具有较低的关联度和较强的分类能力;然后再结合属性与类变量及属性间的相关性对各属性进行加权;最后对待判样本进行分类.经实验结果表明,该算法有效地提升了朴素贝叶斯的分类性能.

Naive Bayes is simple and efficient in handling classification problems.Generally,it assumes that the attributes are conditionally independent,and the degree of influence of each attribute variable on the class variable is the same,but in practical ap-plications,these are difficult to be satisfied,which makes the Classification performance reduced.Therefore,this paper proposes a weighted naive Bayes classification algorithm based on attribute reduction.The algorithm first removes irrelevant attributes and redun-dant attributes according to the classification ability of different values of various attributes and the size of symmetric uncertainty be-tween attributes,so that the filtered attributes have low correlation and strong classification ability.And then it combines the attributes and the correlation between class variables and attributes to weight each attribute,and finally classifies the judgment samples.The ex-perimental results show that the algorithm effectively improves the classification performance of Naive Bayes.