Fine tuning the tree augmented Na?ve Bayes (FTTAN) learning algorithm

In this work, we adapt the fine tuning algorithm of Naïve Bayes (NB) for Tree Augmented Naïve Bayes (TAN). The adapted algorithm, takes into consideration the differences in structure between NB and TAN. The algorithm augments the regular TAN learning phase with a fine tuning phase in which the probability terms are fine tuned to give better classification accuracy. The fine tuning algorithm is applied on five models of TAN: TAN search, K2 search, tabu search, Hill Climber search, and Repeated Hill Climber search. Our empirical results show that fine tuning TAN significantly improves the average classification accuracy of all TAN models in many domains.