Arranging a hybrid-weight for attribute in weighted naïve Bayesian classifier

In this paper, a modified naïve Bayesian classifier with hybrid-weight (NBCH) is proposed. NBCH arranges a weight for each condition attribute by considering the gain ratio and correlation coefficient. The gain ration is used to measure the effectiveness of a condition attribute in the classification task. And, the correlation coefficient is designed to depict the linear dependence between condition attribute and decision attribute. Our strategy calculates the hybrid of gain ration and correlation coefficient and uses this hybrid as the weight of given condition attribute. In order to validate the feasibility and effectiveness of NBCH, we experimentally compare our method with standard naïve Bayesian classifier (NBC), NBC with gain ratio weight (NBCGR), and NBC with correlation coefficient weight (NBCCC) on 10 UCI datasets. And, a statistical analysis is also given. The final results show that NBCH can obtain the statistically best classification accuracy.