Artificial Intelligence in Building Maintenance Performance: A Systematic Literature Review

Abstract

Building maintenance constitutes a substantial proportion of lifecycle expenditure in facility management, where conventional reactive approaches often result in increased operational costs, unexpected system failures, and suboptimal performance outcomes. In response to these limitations, artificial intelligence (AI) and machine learning (ML) have emerged as promising technologies for enabling predictive maintenance in building systems. This study aims to systematically review and meta-analyze existing research on AI-based building maintenance performance prediction, with particular emphasis on identifying key system performance parameters that influence failure patterns. A systematic literature review was conducted following PRISMA guidelines, covering publications between 2005 and 2025 across multiple databases. Inclusion criteria were restricted to journal articles focusing on AI/ML applications in building maintenance prediction. Data extraction encompassed study characteristics, AI techniques, performance metrics, and key empirical findings. A total of 47 studies met the inclusion criteria, representing 15,847 building systems across diverse domains. The analysis indicates that neural networks (32%), random forest (24%), and support vector machines (19%) are the most frequently applied methods, with HVAC systems (45%) and electrical systems (28%) being the dominant application areas. Meta-analysis results reveal a pooled prediction accuracy of 89.3% (95% CI: 87.1–91.5%) for fault detection and a root mean square error (RMSE) of 2.47°C (95% CI: 2.12–2.82°C) for performance prediction. These findings demonstrate that AI-based approaches achieve high predictive accuracy across building systems, with neural networks and ensemble methods showing superior performance in complex environments. Nevertheless, current studies remain largely system-specific and fragmented. Future research should therefore prioritize multi-system integration and real-time implementation to enhance the practical applicability of AI-driven predictive maintenance in facility management.