VOGA: Variable Ordering Genetic Algorithm for Learning Bayesian Classifiers

Abstract

This work proposes a hybrid approach to help the process of learning a Bayesian Classifier (BC) from data. The proposed method named VOGA (and its variant VOGA+) uses a Genetic Algorithm to optimize the BC learning process by means of the identification of an adequate variables ordering. The main contribution of VOGA and VOGA+ is the use information about the class variable when defining the most suitable variable ordering. Trying to optimize the GA initial population, VOGA+ ranks the attributes based on the class variable. Experiments performed in a number of datasets revealed that both methods are promising and VOGA+ tends to be favored domains having higher number of variables.