Machine learning prediction of estimate at completion for infrastructure projects in Nusantara Indonesia

Accurate forecasting of Estimate at Completion (EAC) is essential for monitoring large infrastructure projects, particularly when project conditions change during execution. Conventional Earned Value Management (EVM) formulas estimate future cost through fixed relationships between cost and schedule indicators, which may produce unstable predictions when project performance fluctuates across reporting periods. This study evaluates the use of machine learning (ML) models to support EAC forecasting using project monitoring data. Three prediction models were examined, namely Linear Regression as a baseline statistical model, as well as Neural Network and Deep Learning as nonlinear models that capture complex relationships among monitoring indicators. The models were trained using financial and progress variables derived from e-monitoring records, the performance of which was evaluated using Mean Absolute Percentage Error (MAPE) that represents deviation between model outputs and the reference forecast values used in the monitoring system. The results show differences in deviation across the models, where the Neural Network produced the lowest deviation with a MAPE of 2.26%, while Linear Regression and Deep Learning produced larger deviations. The trained model was then applied to an active road project in Nusantara, Indonesia’s new capital city, where it predicted an EAC of IDR 666 billion compared with the Budget at Completion of IDR 811 billion. These findings indicate that ML models reproduce forecast patterns embedded in monitoring data and generate responsive forecast signals that follow variations in monitoring indicators.