Diversity for Dealing with Drifts (DDD) [1] uses four ensemble classifiers with high and low diversity, before and after a concept drift is detected. A previous study [2] analyzed how these ensembles behaved in data sets suffering from abrupt and gradual concept drifts with several speeds of change, right after the drift and longer after. With the results obtained, ddd was proposed, trying to select the best ensemble (or weighted majority of ensembles) before and after drifts, detected by the use of a drift detection method. The corresponding codes can be downloaded via the links below:
[1] Minku, L.L., Yao, X.: DDD: A new ensemble approach for dealing with concept drift. IEEE Transactions on Knowledge and Data Engineering 24(4) (April 2012) 619-633
[2] Minku, L.L., White, A.P., Yao, X.: The impact of diversity on online ensemble learning in the presence of concept drift. IEEE Transactions on Knowledge and Data Engineering 22(5) (May 2010) 730-742
Silas Garrido - sgtcs@cin.ufpe.br