Improving Fast Adaptive Stacking of Ensembles

The treatment of large data streams in the presence of concept drifts is one of the main challenges in the fields of machine learning and data mining. This article presents two ensemble algorithms designed to quickly adapt to concept drifts, both abrupt and gradual. Fast Stacking of Ensembles boosting the Best (FASEB) and Fast Ensemble boosting the Best with Combined Weighting Voting (FASEB_wv) are adaptations of Fast Adaptive Stacking of Ensembles (FASE), designed to improve their execution time without causing considerable losses in the algorithm's accuracy. In order to achieve a more efficient model, adjustments were made in the update strategy and voting procedure of the ensemble. To evaluate the proposals against the state of the art, we used Naive Bayes (NB) and Hoeffding Tree (HT) as learners to compare the performance of the algorithms on artificial and real-world data sets. The experiments showed that FASEB and FASEB_wv are more efficient than FASE in many scenarios, often also achieving better results in accuracy.