Intelligent Model Management: Using Reinforcement Learning to Detect Data Drift and Retrain Industrial Machine Learning Systems

In the evolving landscape of industrial processes, maintaining the accuracy of machine learning (ML) models is a formidable challenge due to the phenomenon of data drift. This paper introduces an innovative reinforcement learning (RL) framework designed to monitor and adaptively retrain ML models in response to data drift, ensuring sustained model performance in industrial settings. We apply our methodology to a real-world scenario within the cement milling industry, demonstrating that our RL agent can effectively identify optimal retraining intervals and maintain high model accuracy over time. Our results indicate that ML models managed by the RL agent surpass the performance of static models, highlighting the potential of RL in continuous model management. The task-agnostic nature of our approach also suggests its applicability across various domains, paving the way for future research in adaptive systems and realtime data processing.