| Popis: |
Modern machine learning algorithms must be able to fast consume data streams, maintaining accurate results, even with the presence of concept drift. This work proposes AFXGB, an Adaptive Fast binary classification algorithm using XGBoost, focusing on the fast induction of labeled data streams. AFXGB uses an alternate model training strategy to achieve lean models adapted to concept drift. We compared AFXGB with other data stream classifiers using synthetic and real datasets. The results showed that AFXGB is four times faster than ARF and 22 times faster than AXGB, maintaining the same accuracy and with the fastest recovery from concept drifts, thus preserving long-term accuracy. |
