Towards Interpretable Thermal Load Predictive Models: A Systematic Review

Buildings are major consumers of energy; therefore, making accurate thermal load predictions is essential for improving operational efficiency and annual planning. Although machine learning (ML) has enhanced predictive capabilities, most models remain difficult to understand and explore by engineers and other stakeholders. This research presents a systematic review focusing on the role of interpretable models, specifically rule-based classifiers, decision trees, and post hoc feature attribution and interaction approaches such as Explainable AI (XAI), as design-relevant approaches for energy efficiency optimisation. Through screening and analysis, 39 of the 138 identified studies were retained. The review synthesises evidence across several key dimensions such as building envelope and material factors, spatial and geometric design, building typology, climate adaptability, occupancy behaviour, sustainable building system integration, and other related themes. The major contribution is a practice-oriented approach grounded in these themes, thus demonstrating how interpretable models reveal physical dependencies, expose model-to-reality gaps, and improve applicability across diverse building contexts. The findings highlight the need for models that are context-sensitive, easy to understand, cost-effective, and modifiable to guide engineers, designers, managers, and operators in developing thermal load modelling practices.