基于R-vine Copula理论的改进朴素贝叶斯分类器

Improved Naive Bayes Classifier Based on R-vine Copula Theory

朴素贝叶斯分类方法在分类时,基于属性之间的独立性假设影响了其分类性能。基于此在R-vine Copula理论的基础上,利用一系列Pair Copula函数和核密度函数的乘积来构造属性的类条件概率密度函数,并通过AIC准则选取最合适的Pair Copula函数,用极大似然估计法确定其参数。实验结果表明,改进的分类器提高了分类的准确率,避免了因属性相关导致的分类效果的偏差。

When the naive bayesian classification method classifies,the independence assumption between its attributes affects its classification performance.Based on the R-vine Copula theory,this paper constructs a class-conditional probability density function by using a product of a series of Pair Copula functions and a kernel density function,and it selects the most suitable Pair Copula function by AIC criterion,whose parameters are determined by maximum likelihood estimation method.The experimental results show that the improved classifier boosts the accuracy of classification and avoids the bias of classification effect caused by attribute correlation.