Online bagging of evolving fuzzy systems

Evolving fuzzy systems (EFS) have received increased attention from the community for purpose of data stream modeling in an incremental, single-pass and transparent manner. To date, a wide variety EFS approaches been developed successfully used real-world applications which address structural evolution parameter adaptation single models. We propose specific ensemble scheme to increase their robustness predictive performance on new samples. Our approach relies online variant bagging various members are generated bags, that is, updated based probabilistic sampling technique, this with guaranteed convergence classical batch bagging. The autonomous pruning is undertaken omit undesired atypically higher errors than other members. two variants, hard where deleted forever ensemble, soft receive weights calculate overall prediction, according performance; thus, who at certain point time may be dynamically recalled later stage. carried out whenever drift detected, significantly worsening indicator, measured terms Hoeffding inequality. Newer typically represent drifted state better thus up-weighed compared older within advanced (weighted) calculation prediction. termed bagged (OB-EFS) was evaluated models related SoA four streams (containing noise levels, drifts operating conditions) showed lower prediction error trend lines.