<p>Most real-world events generate large quantities of data that are processed in real time, known as data streams. These data streams are dynamic, uninterrupted, and continuous. The evolving nature of streaming data can eventually lead to changes in the underlying distributions, a phenomenon known as concept drift. Due to data drift, online learning models may experience a significant drop in performance. In this paper, we propose an Error-Weighted Drift-Adaptive Ensemble Framework (EW-DA-EF) for the classification of drifting data streams. In EW-DA-EF, the online ensemble model Adaptive Random Forest (ARF), constructed from Hoeffding Trees, is integrated with five popular drift detection techniques (EDDM, DDM, ADWIN, HDDM_A, and HDDM_W) to construct five base learners: ARF-EDDM, ARF-DDM, ARF-ADWIN, ARF-HDDM_A, and ARF-HDDM_W. The weights of the five base learners are dynamically updated based on the real-time relative error rates of individual base learners during the online learning process. We observe from results on both real-world drifting datasets (weather and LUdata) and synthetic drifting datasets (sea_stream and sea_big) that EW-DA-EF shows superior performance in terms of accuracy, precision, recall, and F1-score as compared to the state of the art. The inclusion of five different drift detection techniques in the ensemble ensures effective adaptation to both gradual and abrupt types of drift.</p>