摘要：In this study, we introduced a novel feature ranking technique applicable to two well known classifiers Bayesian Belief network and Random Forest as both of these classification systems have been shown to be sensitive to the initial ordering of the features. We have illustrated that improvement in classification can be obtained even without ceding variables for feature (attribute) ranking sensitive classifiers. We also performed a comparison between Bayesian Belief network and Random Forest classification approaches in the well known feature subset selection and feature ranking problem. The proposed technique Polarization Measure (herein known as PM) is originated from within joint probability to discover the degree of explanation made by first feature (attribute)’s state to explain the other feature’s state. The technique has significantly better well performed in Bayesian belief network and better in random forest classifier in comparison to five feature ranking techniques and three well established feature subset selection techniques
关键词：random forests algorithm; machine learning; Bayes structure learning; ranked features
