Dynamic Ensemble of Ensembles in Nonstationary Environments

Abstract

Classifier ensemble is an active topic for learning from non-stationary data. In particular, batch growing ensemble methods present one important direction for dealing with concept drift involved in non-stationary data. However, current batch growing ensemble methods combine all the available component classifiers only, each trained independently from a batch of non-stationary data. They simply discard interim ensembles and hence may lose useful information obtained from the fine-tuned interim ensembles. Distinctively, we introduce a comprehensive hierarchical approach called Dynamic Ensemble of Ensembles (DE²). The novel method combines classifiers as an ensemble of all the interim ensembles dynamically from consecutive batches of non-stationary data. DE² includes two key stages: (1) Component classifiers and interim ensembles are dynamically trained; (2) the final ensemble is then learned by exponentially-weighted averaging with available experts, i.e., interim ensembles. We engage Sparsity Learning to choose component classifiers selectively and intelligently. We also incorporate the techniques of Dynamic Weighted Majority, and Learn^+ +.NSE for better integrating different classifiers dynamically. We perform experiments with the data in a typical non-stationary environment, the Pascal Large Scale Learning Challenge 2008 Webspam Data, and compare our DE² method to other conventional competitive ensemble methods. Experimental results confirm that our approach consistently leads to better performance and has promising generalization ability for learning in non-stationary environments.