A machine learning and MCDM framework for the data-informed sustainable project scheduling and material procurement problem

Abstract

Construction project performance is strongly influenced by the coordination of scheduling, material procurement, and supplier selection decisions. However, these decisions are often addressed separately, despite their significant interdependence and their joint impact on project cost, completion time, and supply chain sustainability. This study proposes an integrated framework that combines machine learning (ML), multi-criteria decision-making, and multi-objective optimization to support construction project scheduling and supplier selection under material ordering decisions. To the best of our knowledge, this is the first study to employ ML within integrated project scheduling and material ordering decisions for predicting both activity duration and activity execution cost. In the first stage, ML models are developed to predict activity duration and execution cost based on historical project data, and the results show that XGBoost outperforms Random Forest, achieving $R^2$values of 0.901 and 0.923, respectively. In the second stage, supplier evaluation is conducted using multiple criteria, including price, lead time, reliability, quality, capacity, and ESG performance, through a structured decision-making process. In the final stage, the predicted activity parameters and supplier information are integrated into a multi-objective optimization model to minimize total project cost and project duration while maximizing supplier ESG performance. The Augmented Epsilon Constraint method is employed to generate Pareto-optimal solutions. Results reveal clear trade-offs among objectives, demonstrating that ordering and supplier decisions vary significantly depending on managerial priorities. Overall, the proposed framework improves predictive accuracy and provides a practical decision-support tool for more integrated and sustainability-aware construction planning.