Deep Learning Model Performance for Construction Cost Prediction in Main Contracting Organizations: A Systematic Review and Meta Regression Analysis

Accurate cost prediction of construction projects has become a serious issue in the civil engineering field. This systematic literature review synthesizes latest developments in cost estimation models, specifically pertaining to the fields of soft computing and machine learning applications in construction cost management. 52 peer-reviewed articles covering more than 55 independent datasets published between 2014 and 2024 were systematically located in Scopus, Web of Science, and Google Scholar. The conducted meta-regression analysis showed that hybrid and ensemble models were significantly more efficient than singular algorithms, with an average 8-15% reduction in Mean Average Percentage Error when compared to traditional statistical techniques. These performance results translate into improved early-stage budgeting accuracy and enhanced reliability of cost predictions in contractor-led project decision environments. The review further identifies key practical constraints influencing real-world adoption, including data availability, model interpretability, and organizational readiness, while highlighting opportunities for integration of explainable hybrid models within existing cost-management workflows. Overall, the findings provide actionable insights for advancing standardized, robust, and industry-deployable cost estimation systems in civil engineering practice.