Weighted Naïve Bayesian classification algorithm based on correlation coefficients

Naïve Bayesian classifier is one of the most effective and efficient classification algorithms, but its conditional independence assumption affects its classification performance. Correlation coefficients is used to overcome this impractical assumption by proposing a new algorithm called weighted Naïve Bayes based on correlation coefficient (WNB-CC). By computing correlation coefficients between each conditional attribute and a decision attribute, different conditional attributes are weighted differently. WNB-CC is experimentally tested in terms of classification accuracy, using the ten data sets from the UCI machine learning repository, and compare it to Naïve Bayes (NB), rough set (RS), weighted Navie Bayes based on hill climbing (WNB-HC) and C4.5. The experimental results show that our WNB-CC can improve the classification performance.