Cluster-Specific Localized Drift Detection for Efficient Batch Model Adaptation under Controlled Distribution Shift

This work presents a cluster-induced distribution shift simulation framework that enables the transformation of static tabular datasets into controlled evolving data streams, facilitating the evaluation of drift adaptation methods. Six adaptation strategies, including static learning and various retraining approaches, were assessed across five benchmark datasets for both classification and regression tasks using multiple predictive model families. This framework is significant for practitioners as it provides a structured methodology to evaluate and improve model robustness in dynamic environments where data distributions change over time.