Demystifying machine learning for predictive analytics in construction

This is a horizontal analysis on the state of research of machine learning (ML) for construction applications. The objective is to identify relevant topics in the research area and clarify the actual capabilities and limitations of ML approaches for construction. A literature review and thematic analyses were conducted to identify significant topics as well as an analysis of the most cited papers and use cases. A discussion of relevant applications and challenges is presented as well. The key findings are (1) there has been a massive increase on research efforts in the area; however, research is still behind in the use of state-of-the-art models, such as large language models, transformers, and reinforcement learning. Most importantly, it is usually limited to the use of relatively small datasets. (2) There are still significant challenges regarding the creation of sufficiently large datasets, but there are effective manners to address those challenges including the creation of synthetic data. This study provides construction practitioners and researchers with an overview of the key aspects of research on ML in construction that will help improve the understanding in this research area.