A weighted flexible naive Bayesian classifier for continuous attributes

In this paper, a weighted flexible naive Bayesian classifier (simply WFNB) is proposed to handle the classification problem with the continuous attributes. In WFNB, every marginal probability density function (p.d.f.) in the class-conditional p.d.f. is assigned a weight by measuring the dependence between the corresponding condition attribute and the class attribute. The dependence is the mutual information between the individual attribute and the class attribute. Then, we compare the classification accuracy of WFNB with the traditional flexible naive Bayesian classifier (FNB) on 10 VCI datasets. The experimental results show that our proposed WFNB can obtain the statistically better classification accuracy and demonstrate that WFNB is feasible and efficient.