A Novel Approach of Adpative Window 2 Technique and Kalman Filter- “KalADWIN2” for Detection of Concept Drift

A recommendation engine (RE) is a machine learning technique that provides personalized recommendations and anticipates a user's future preference for a collection of goods or services. In Online Supervised Learning (OSL) settings like various REs, where data vary over time, Concept Drift (CD) issue usually occurs. There are many CD Detectors in the literature work but the most preferred choice for the non-stationary, dynamic and streaming data is the supervised technique- Adaptive Window (ADWIN) approach. The paper aims towards the limitations of the ADWIN approach, where ADWIN2 approach is more time &memory efficient than ADWIN. The paper also focusses on novel proposed technique of the combination of Kalman Filter and ADWIN2 approach, named-“KalADWIN2”, as it’s the best estimator for detection even in noisy environment. It ultimately helps in fast CD detection in REs.