Cluster-Specific Localized Drift Detection for Efficient Batch Model Adaptation under Controlled Distribution Shift
For practitioners deploying ML models in dynamic environments, this work provides a reproducible evaluation framework and demonstrates that localized drift detection can improve adaptation efficiency.
The paper proposes a cluster-induced distribution shift simulation framework to convert static tabular datasets into controlled evolving data streams, and evaluates six adaptation strategies, finding that cluster-local ADWIN retraining offers efficient batch model adaptation under controlled distribution shift.
Machine learning systems deployed in dynamic environments frequently operate under nonstationary data distributions, where controlled distribution shift can progressively degrade predictive performance. However, many widely used tabular benchmark datasets lack explicit temporal structure, limiting reproducible evaluation of drift adaptation methods. This work proposes a cluster-induced distribution shift simulation framework that transforms static tabular datasets into controlled evolving data streams through structured perturbations across featurespace partitions. Using this framework, six adaptation strategies are systematically evaluated: static learning, sliding-window retraining, global ADWIN retraining, cluster-local ADWIN retraining, random subspace drift detection, and feature-partitioned drift detection. Experiments are conducted on five benchmark datasets covering both classification and regression tasks using diverse predictive model families, including linear models, k-Nearest Neighbours, tree ensembles, boosting methods, and adaptive online learners.