A Novel Multiple Classifiers Integration Algorithm with Pruning Function

For improving identification rate and real time of ensembles learning algorithm, the diversity of ensemble classifiers is analyzed and a novel combination algorithm with pruning function of multiple classifiers is presented. A coincident errors measure of classifiers is presented for the compound error probability by which classifiers are partitioned, and some classifiers in a partition are pruned. The voting weights of pruned classifiers are assigned according to diversity between classifiers, so that optimize classifier set and voting weights for integration are obtained. The UCI data depository and Radar Radiant Point data are used as test data, and the result of experiment show that classifiers ensemble with pruning can get similar classification accuracy as accuracy of entire classifier integration and reduce classification time.