Machine Learning‐Enabled Occupancy Modelling for Energy‐Efficient and Sustainable Building Operations: A Critical Review of Strategies and Future Directions

ABSTRACT The study of occupancy behaviour (OB) and prediction of building energy consumption is an important topic, while machine learning (ML) with the use of real‐time data and adaptive algorithms has become more relevant. Although research in this area is rapidly expanding, each of the models developed so far still presents limitations in scalability and transferability to various building environments attributed to the differences in modelling strategies, data inputs, sensing approaches and evaluation methods. The current review studies have tended to be limited to a specific aspect of occupancy modelling (algorithm, sensing technology or control strategy) and a more comprehensive synthesis is needed that incorporates these in the broader context of energy‐efficient building operations. This review aims to critically discuss recent approaches for modelling occupancy behaviour and predicting energy consumption using ML techniques and identifies their modelling approaches, modes of sensors, context of applications and methods of evaluation. Over 80 peer‐reviewed studies were critically reviewed to formulate a structured synthesis of existing modelling approaches, compare their applications across different building types and identify methodological advancements and remaining challenges that hinder practical deployment. The results of the analysis suggest a strong trend towards non‐intrusive sensing, hybrid and context‐aware modelling, explainable and privacy‐preserving artificial intelligence and the growing use of sensor fusion to enhance the reliability of the predictions. Meanwhile, there is a lack of model transferability, benchmarking inconsistencies, data heterogeneity and a lack of standard datasets, which hinders broader deployment. This review succinctly summarizes existing research and points out future research directions that will enable the construction of scalable, transferable and ethically responsible occupancy‐aware energy management systems and can serve as a handy resource for researchers, engineers and policymakers. The results also show how the development of occupancy modelling leverages ML can support the United Nations Sustainable Development Goals (SDGs), notably SDG 7 (Affordable and Clean Energy), SDG 11 (Sustainable Cities and Communities) and SDG 13 (Climate Action).