An Integrated Machine Learning Framework for Predictive and Sustainable Road Construction Project Management

Abstract

Road construction projects are often affected by cost overruns, schedule delays, technical uncertainty, and coordination challenges, making proactive project control difficult. This study proposes an integrated machine learning framework for predictive and sustainable road construction project management. The framework uses structured project data to predict key outcomes, including actual cost, actual duration, risk level, and completion percentage. It combines target-specific models, overrun-based target engineering, logical output constraints, and an interpretable explanation layer.

The framework was developed using a dataset of 1000 road construction projects and evaluated on 100 holdout projects. For visual clarity, a representative 10-project sample is presented in the results section, while the reported performance metrics are based on the full holdout set. The results show strong predictive performance for the main targets. Cost prediction achieved a MAPE of 2.89% and an R² of 0.99, while duration prediction achieved a MAPE of 3.09% and an R² of 0.97. Risk classification reached an accuracy of 90%. Completion percentage showed comparatively weaker performance and is therefore treated as a supporting indicator.

Overall, the findings show that the proposed framework can provide accurate, interpretable, and practically useful support for road project forecasting. The study contributes to intelligent construction management by offering an integrated decision-support framework for more proactive and sustainable infrastructure delivery.