Akademik Veri Yönetim Sistemi
JOURNAL OF INDUSTRIAL AND MANAGEMENT OPTIMIZATION, cilt.22, sa.4, ss.1928-1955, 2026 (SCI-Expanded, Scopus)
Accurately predicting construction labor productivity is essential for effective project management, particularly in resource-constrained settings like Somaliland. However, existing studies primarily rely on single-model structures, which often struggle to generalize across varying site conditions and activity types. Moreover, while metaheuristic optimization has been used to tune individual models, its integration into ensemble architectures remains limited, leaving a gap in methods capable of combining complementary learning behaviors. To address this, this study developed a stacked ensemble model designed to improve prediction stability and robustness. The proposed approach integrates two metaheuristically optimized Random Forest models, tuned using Particle Swarm Optimization and Artificial Bee Colony, with an Extreme Gradient Boosting (XGBoost) meta-learner. The model was evaluated on 1422 real-world productivity records from 52 construction projects involving concrete, formwork, and bricklaying tasks. The stacked ensemble achieved competitive and more stable performance, particularly for high-variance activities, demonstrating improved robustness, stability, and generalization. Feature selection results emphasized the importance of safety-related disruptions, weather conditions, and communication efficiency in determining labor productivity. The proposed model offers a practical and scalable tool for more reliable construction labor productivity (CLP) prediction and provides decision-making support for construction management, particularly in resource-limited environments.