Generalized additive Bayesian network classifiers

摘要: Bayesian network classifiers (BNC) have received considerable attention in machine learning field. Some special structure BNCs been proposed and demonstrate promise performance. However, recent researches show that BNs may lead to a non-negligible posterior problem, i.e, there might be many structures similar scores. In this paper, we propose generalized additive classifiers, which transfers the problem models (GAM) problem. We first generate series of very simple BNs, put them framework GAM, then adopt gradient-based algorithm learn combining parameters, thus construct more powerful classifier. On large suite benchmark data sets, approach outperforms traditional BNCs, such as naive Bayes, TAN, etc, achieves comparable or better performance comparison boosted classifiers.