ENHANCEMENT OF BAYESIAN MODEL WITH RELEVANCE FEEDBACK FOR IMPROVING DIAGNOSTIC MODEL

Abstract

An enhanced method to classify multi-class clinical disease is proposed in this study. The enhanced method is based on the Bayesian Model, which incorporates Bayes’ rule and probability theory. It covers three main components: prior, conditional, and posterior probability. The recommended enhancement method is the Bayesian Relevance Feedback (BRF) Model. BRF can solve the non-existent value of posterior probabilities (zero values of probability), focusing on increasing the classification accuracy in the diagnosis of disease. The BRF has the capability to produce significant classes or target (cancer stage) by exploiting relevance feedback. Consequently, models based on eight different classifiers—K-Nearest Neighbors, Bayesian Model, Rule OneR, Meta MultiClass Classifier, Multilayer Perceptron, Random Tree, SMO-Poly Kernel, and Naive Bayes—were applied in the evaluation process. The results of the experimental works using an oral cancer dataset show that BRF outperformed the eight other classifier models, achieving 95.83% classification accuracy.