A Fuzzy Logic Ensemble Approach to Concept Drift Detection.

Concept drift occurs when the statistical properties of a data distribution change over time, causing the performance of machine learning models trained on prior data to degrade. This is a prevalent issue in many real-world applications where the data distribution can shift due to factors such as user behaviour alterations, environmental changes, or modifications in the data-generating system. Detecting concept drift is crucial for developing robust and adaptive machine learning systems. However, identifying excessive drifts may lead to decreased model performance. In this article, we present an ensemble method employing multiple concept drift detectors to detect concept drift. A fuzzy logic approach balances the outputs of the various drift detectors comprising the ensemble. The proposed framework can handle concept drift in regression problems. Experimental results demonstrate enhanced efficiency in detecting concept drifts under different conditions.