Coping with Data Distribution Shifts: XAI-Based Adaptive Learning with SHAP Clustering for Energy Consumption Prediction

Adapting to data distribution shifts after training remains a significant challenge within the realm of Artificial Intelligence. This paper presents a refined approach, superior to Automated Hyper Parameter Tuning methods, that effectively detects and learns from such shifts to improve the efficacy of prediction models. By integrating Explainable AI (XAI) techniques into adaptive learning with SHAP clustering, we generate interpretable model explanations and use these insights for adaptive refinement. Our three-stage process: (1) SHAP value generation for the model explanation, (2) clustering these values for pattern identification, and (3) model refinement based on the derived SHAP cluster characteristics, mitigates overfitting and ensures robust data shift handling. We evaluate our method on a comprehensive dataset comprising energy consumption records of buildings, as well as two additional datasets, to assess the transferability of our approach to other domains, regression, and classification problems. Our experiments highlight that our method not only improves predictive performance in both task types but also delivers interpretable model explanations, offering significant value in dealing with the challenges of data distribution shifts in AI.