Decision by maximum of posterior probability average with weights: a method of multiple classifiers combination

In this paper, a new multiple classifiers combining algorithms, that is maximum of posterior probability average with weight (MAW rule), is introduced. We adopt the same methods with Bagging to train single classifier in this algorithm, but we amend integration rule, which the result lies on maximum in average with weight for every class rather than majority vote. This algorithm bases on parallel integration, and naive Bayesian classification is used to construct single classifier. Besides our combining algorithm, we also select other algorithms, which are Max rule, Min rule, Majority vote rule, Sum rule, and Product rule as comparing objects. According to experiments on KDD99 dataset and the letter dataset of UCI, MAW rule lead to less error than other combining algorithms and better performance.