A Novel Explainable Machine Learning Decision- Support Framework for Residential Building Energy Consumption Prediction and Management

Accurate prediction of electrical energy consumption is a fundamental requirement for intelligent energy management systems, demand response programs, and residential building energy optimization. Although recent machine learning algorithms have significantly improved prediction accuracy, most previous studies have primarily emphasized predictive performance while paying limited attention to model explainability and the transformation of prediction outputs into practical decision-making solutions. This study proposes an explainable machine learning-based intelligent decision-support framework that integrates data preprocessing, domain knowledge-based feature engineering, model training and evaluation, SHAP-based explainability analysis, recommendation generation, and external validation within a unified architecture. Five machine learning algorithms Linear Regression, Random Forest, XGBoost, LightGBM, and CatBoost- were systematically evaluated using standard performance metrics. Although Linear Regression achieved the best quantitative performance on the primary dataset, CatBoost was selected as the core model of the proposed framework because it provided competitive predictive performance together with comprehensive explainability, successful external validation, and effective support for decision-making. The SHAP method was employed to investigate the factors influencing predictions at both the global and local levels. The extracted explanations were subsequently transformed into practical recommendations through the proposed decision-support engine. External validation using the independent REFIT House ۱ dataset further demonstrated the satisfactory generalization capability of the proposed framework. To improve transparency, reproducibility, and future research development, the source code, technical documentation, research notebooks, and principal outputs have been released through GitHub, while a stable version has been archived in Zenodo with a permanent DOI. The proposed framework extends beyond conventional energy consumption prediction by integrating prediction, model explainability, and decision support within a unified architecture, thereby providing an effective platform for intelligent residential energy management.

کلیدواژه ها:

Energy Consumption Prediction ، Machine Learning ، Explainable Artificial Intelligence ، SHAP ، Feature Engineering ، CatBoost ، Decision Support System ، Intelligent Energy Management