Divide-and-Conquer–Based Stratified Models for Predicting Precast Concrete Installation Times

Precast concrete (PC) is increasingly recognized for offering sustainable building solutions in urban development. However, the realization of its benefits crucially depends on strategic planning and coordination throughout the entire process, from production to installation. This study is part of efforts to employ data-driven approaches to predict PC installation times. Specifically, this study employs a divide-and-conquer approach to stratify the PC installation process into two levels of detail, develops a distinct model for each level, evaluates these models under four regression algorithms, and determines the level of detail and the algorithms that are most suitable for creating reliable predictive models. Utilizing data collected from a logistics center in South Korea, this study found that the random forest model significantly outperformed others in forecasting the PC installation process. Additionally, the nine-stage decomposed model outperformed the two-stage aggregate model, particularly through enhanced feature selection tailored to the complexities of each work step. The study also identified significant variability in prediction accuracy among different work steps, with installation-related work steps showing more reliable predictions than unloading-related work steps. These refined predictions can support PC project managers by providing insights that identify time-intensive work steps and optimize planning to minimize delays, thereby enhancing overall project efficiency.