Machine learning techniques for predictive maintenance of building services: a comprehensive review and research outlook

Purpose: Predictive maintenance in buildings is crucial for minimising unplanned downtime and extending lifespan of building components, yet its implementation remains complex. Machine learning (ML) offers a transformative approach by enabling systematic predictions and automation. The purpose of this study is to analyse the interrelationship between ML techniques and predictive maintenance of building services, identifying key research trends and future directions. Design/methodology/approach: A bibliographic analysis was conducted on 118 journal articles using VOSviewer to examine co-authorship and co-occurrence patterns. The key themes generated were then explored semi-systematically, focusing on the most frequently used ML techniques and predictive maintenance applications. Findings: The results reveal a strong relationship between ML and predictive maintenance, with increasing research interest post-2021. Co-occurrence analysis highlights the evolution of research themes, shifting from conventional ML models to advanced techniques such as digital twins and lifelong learning with deep generative replay modelling. Among the most frequently applied ML techniques, Extreme Gradient Boosting, Artificial Neural Networks and Deep Neural Networks have demonstrated the best predictive performance in fault diagnostics and system optimisation. Practical implications: The findings advocate for stronger interdisciplinary collaborations among researchers, institutions and industries to bridge the gap between research advancements and real-world implementation in facilities management and building life cycle. Originality/value: This study provides a comprehensive examination of research trends, highlighting underexplored ML applications in building services predictive maintenance.