Parameter Free and Non Penalized Scoring Metric for Bayesian Belief Network

Abstract

This study has introduced parameter free, decomposable with no penalty factor and an efficient saleable scoring metric, the Non Parametric Factorized Likelihood Metric (NP-FiLM) useful for structure learning. The proposed score metric has its root in information theoretic elucidation. The metric is devised to maximize the discriminant function for query variables with respect to the class and other non class variables. An empirical evaluation of the proposed metric has been carried out over an abundant number of natural datasets obtained from UCI. The comparison is made with respect to eleven tree classifiers, one regression model and two neural network system. Furthermore, the scoring metric has been examined to six peer scoring metrics within the greedy search mechanism. NP-FiLM oriented Bayesian Belief Network have been satisfactory found with significant results in a paradigm of accuracy and classification error. The introduced scoring function is capable of illustrating the best possible data fitting in the context of hyper-parameters described above.