An ant colony optimization-based random forest model for estimating labor productivity of concrete activities in Egyptian construction sites

Accurate labor productivity forecasting is essential for reliable resource planning, cost control, and schedule management in construction projects. However, activity-level productivity prediction remains challenging because site productivity is affected by nonlinear interactions among activity type, working height, weather conditions, overtime, crew size, and achieved quantity. This study develops an ant colony optimization-based random forest model (ACO-RF) for estimating labor productivity in concrete-related construction activities. The dataset was systematically compiled from daily site records of Egyptian reinforced-concrete building projects and comprised 2,062 valid daily operation records. The model was applied to nine activity groups: column formwork, column reinforcement, column concreting, slab formwork, slab reinforcement, slab concreting, wall core formwork, wall core reinforcement, and wall core concreting. ACO was used to optimize the RF modelling process through feature selection and hyperparameter tuning, while RF was used to capture nonlinear relationships within the productivity dataset. The proposed model was evaluated using standard error-based and goodness-of-fit metrics and compared with benchmark machine-learning models. The results indicate that ACO-RF achieved the strongest predictive performance among the evaluated models, supporting more reliable activity-level productivity forecasting for concrete works.