ERP-Integrated Cost Value Reconciliation Automation in Construction Main Contracting Organizations Using a Hybrid Machine Learning Framework

• Presents an Enterprise Resource Planning (ERP)-integrated automation framework that embeds machine-learning-based cost forecasting directly within Cost Value Reconciliation (CVR) workflows of construction main contractors. • Empirically evaluates the framework using ERP-derived cost data from 46 UAE construction projects, validated under temporally consistent CVR decision conditions. • Demonstrates that hybrid ensemble models (LightGBM - Random Forest) achieve substantially higher predictive accuracy and stability than conventional spreadsheet-driven CVR approaches (R² improved from 0.33 to 0.80). • Shows that ERP-embedded predictive CVR enables earlier identification of cost deviations and supports more proactive financial control during project execution. Cost Value Reconciliation (CVR) is a core project financial control mechanism used by construction main contracting organizations to monitor cost performance against budgeted targets. However, current CVR implementation remains largely spreadsheet-driven and fragmented across enterprise systems, limiting timely cost deviation identification and proactive financial control. This paper presents an Enterprise Resource Planning (ERP)-integrated, Machine-Learning (ML)-enabled framework designed to automate and enhance CVR-oriented cost forecasting within main contracting organization workflows. Using ERP-derived project cost data from a UAE-based main contractor, the approach is evaluated under temporally consistent validation conditions. Results show that hybrid ensemble models embedded within structured ERP data pipelines demonstrate the potential to enhance predictive stability and decision support within conventional CVR workflows through structured forecasting and uncertainty quantification. Across five iterative refinement stages, predictive performance increased from an R 2 = 0.33 in the initial baseline configuration to R 2 = 0.80 in the optimized LightGBM-Random Forest hybrid configuration. These results demonstrate the feasibility of shifting CVR from retrospective reporting to predictive cost control, enabling earlier detection of cost deviations and more informed financial decision-making during construction project delivery.