An ensemble machine learning approach for measuring and predicting formwork labor productivity

Abstract The construction industry is a major driver of innovation and economic growth. A key factor in this growth is the continuous focus on improving labor productivity, which directly impacts project timelines and costs. Maximizing labor productivity is crucial for overcoming project challenges, accurately forecasting activity durations, and achieving optimal efficiency. Despite its importance, many traditional methods for estimating labor productivity rely on theoretical assumptions that often prove to be inaccurate, leading to significant inefficiencies in both time and cost management. This issue is particularly pronounced in Egypt, where there are no standardized criteria for predicting labor productivity in the construction sector. Recent studies have shown that productivity has a significant effect on project cost and duration, indicating that this issue needs to be studied carefully in Egypt. Therefore, this research addresses the challenges of inaccurate productivity estimation by introducing a novel, data-driven approach. The study focuses on the specific activity of column formwork and was conducted at three Egyptian construction sites: Smouha, Beheira, and Moharam-Bek. The study analyzed 198 records collected from real construction sites. These records were distributed among the three studied sites as follows: 63 records from Smouha, 57 records from Beheira, and 78 records from Moharam-Bek. The results showed that several factors significantly influence construction labor productivity. These factors were ranked using the Relative Importance Index (RII) and were used as independent input variables for the predictive model. Real-time productivity data were collected at two-hour intervals from the three construction sites. This provided a robust dataset to establish baseline productivity values for each location and to quantify time and cost losses. These findings highlight the need to reconsider data collection and development policies to increase their effectiveness in supporting scientific research and enhancing productivity management practices in the construction sector. Furthermore, this model predicts labor productivity for column formwork every two hours, providing a more accurate and reliable tool for project scheduling and decision-making. The model demonstrated excellent performance, achieving a high accuracy of 97.35% and a low validation error of 0.015. In addition, the predictive model was integrated into a website, creating a user-friendly tool for industry professionals. This platform serves as an accurate baseline model that project planners and practitioners can use to forecast labor productivity before starting a project, enabling proactive optimization and cost reduction. Overall, this research represents a significant advancement over traditional estimation methods by offering a precise, data-driven, and practical tool that can substantially improve project planning and execution in the construction industry.