Comparative Analysis of Concept Drift Detection Algorithms for Electrical Streams Across Different Generation Modalities
Idris B. Ismail, Abdul Azeem, Shahwar Shamim, Shahmir Shamim · Apple Academic Press eBooks · 2025
Concept Drift (CD), the shifting nature of data distributions over time, poses a significant challenge in domains like electrical load forecasting, control systems, and the oil and gas industries. This study explores CD detection algorithms in electrical data streams with different generation modalities. We establish a link between CD and electrical data streams, emphasizing the need for adaptive algorithms. The methodology of the study rigorously evaluates various algorithms, including Adaptive Random Forest with Adaptive Windowing (ARF-ADWIN), Adaptive Random Forest with Drift Detection Method (ARF-DDM), Sparse Random Projection with Adaptive Windowing (SRP-ADWIN), Sparse Random Projection with Adaptive Hierarchical Drift Detection Method (SRP-HDDMA), Extremely Fast Decision Tree (EFDT), and Hoeffding Tree (HT). The study employs Symmetric Mean Absolute Percentage Error (SMAPE) as the performance metric. SRP-ADWIN emerges as the standout performer, achieving the lowest SMAPE score of 7.60. This algorithm’s prowess in feature reduction and adaptive windowing proves effective in handling dimensionality and adapting to dynamic data distributions. The insights gained empower industries to make informed, data-driven decisions in real time, even as data dynamics evolve.