Incremental learning based on ensemble pruning

Abstract

Bagging, a widely used ensemble method, is simple and fast, and can generate heterogeneous base classifiers. This research proposes an incremental learning algorithm, PBagging++, based on ensemble pruning. In the algorithm, Bagging is adopted to generate a set of heterogeneous classifiers for each incremental data set. Then an ensemble pruning method is used to select base classifiers from the generated ones and add them to the target ensemble. The new target ensemble will perform the prediction on new instances. Experimental results show that ensemble pruning is an effective way to improve the predictive performance for ensemble based incremental learning.