Abstract This study synthesises applications of machine learning (ML) in construction risk prediction to clarify how current research supports intelligent injury prevention. It examines how ML has been used to predict construction risk, the main methodological and thematic patterns in the literature, and the gaps that limit its contribution to proactive safety management. Bibliographic data were retrieved from Scopus using the keywords “machine learning,” “construction risk,” and “construction safety.” VOSviewer and Bibliometrix were used for keyword co-occurrence, co-authorship, and thematic analyses, interpreted through a socio-technical systems perspective. The findings reveal four dominant research clusters: safety management, predictive modelling, digital technologies, and site-level deep learning. Research is led mainly by the US and China, with notable contributions from the UK, Canada, South Korea, and Hong Kong. The synthesis identifies three key gaps: limited theorisation of how ML outputs inform safety decisions, insufficient attention to worker-centred and organisational factors in model design, and fragmented treatment of data governance and ethical risks. The study proposes a conceptual framework linking data, models, and organisational use.
NKOSI et al. (Thu,) studied this question.