Performance Evaluation of the Tree Augmented Naïve Bayes Classifier for Knowledge Discovery in Healthcare Databases
Resumo
Smart decision support systems (DSSs) have been successfully employed in several areas. In healthcare, these systems offer solutions for uncertain reliably acts and moments. Systems based on Bayesian networks (BNs) can generate predictions even in information lack situations. This paper proposes the modeling and presents a performance evaluation study of the Bayesian classifier named Tree Augmented Na¨ıve Bayes (TAN). Results show that the proposed algorithm obtained good performance for a pregnancy database, presenting F-measure 0.92, Kappa statistic 0.8932, and ROC area 0.993. The proposed method allows representing more complex connections between variables. Nevertheless, it requires major computational effort and time that are not needed in other Bayesian algorithms.