An improving online accuracy updated ensemble method in learning from evolving data streams
Xiaofeng Gu, Jiawen Xu, Shi-Jing Huang, Liao-Ming Wang · 2014
Most stream classifiers need to detect and react to concept drifts, as traditional machine learning goes to big data machine learning. The most popular ways to adaptive to concept drifts are incrementally learning and classifier dynamic ensemble. Recent years, ensemble classifiers have become an established research line in this field, mainly due to their modularity which offers a natural way of adapting to changes. However, many ensembles which process instances in blocks do not react to sudden changes sufficiently quickly, and which process streams incrementally do not offer accurate reactions to gradual and incremental changes. Fortunately, an Online Accuracy Updated Ensemble (OAUE) algorithm was presented by Brzezinski and Stefanowski. OAUE algorithm has been proven to be an effective ensemble to deal with drifting data stream. But, it has a potentially weakness to adaptive to sudden changes as it uses a fixed window. Therefore, we put forward a Window-Adaptive Online Accuracy Updated Ensemble (WAOAUE) algorithm, which is based on OAUE, and a change detector is added to the ensemble for deciding the window size of each candidate classifier. The proposed algorithm was experimentally compared with four state-of-the-art online ensembles, include OAUE, and provided best practice for big data stream mining.