Applied & Interdisciplinary
Hossein Jahangiri; Pooria Rashvand; Ahmad Shokouhfar; Reza Farokhzad; Nima Amani
Abstract
Accurate forecasting of final construction costs is important for project planning and cost control in inflation-sensitive construction markets. This study develops a BIM-integrated artificial neural network (ANN) workflow that links item-level Fehrestbaha data and official construction adjustment indices ...
Read More
Accurate forecasting of final construction costs is important for project planning and cost control in inflation-sensitive construction markets. This study develops a BIM-integrated artificial neural network (ANN) workflow that links item-level Fehrestbaha data and official construction adjustment indices to BIM components and forecasts unit costs at a user-specified completion date. Historical cost information for sixteen materials over 2013--2022 was used to develop the deterministic ANN predictor, and the resulting unit-cost forecasts were returned to the BIM quantity-takeoff workflow. The framework was evaluated on a single nine-story residential building in Tehran by comparing the April 2022 initial estimate, the ANN-based December 2023 forecast, and the actual final project cost. The December 2023 case lies beyond the 2013--2022 historical development horizon and is therefore used as the principal project-level temporal check, whereas the internal training/test diagnostics are interpreted only as model-fit diagnostics because the monthly resampling procedure can induce dependence among adjacent observations. For this case, the absolute project-level deviation was 11.30% for the ANN forecast compared with 23.89% for the initial no-update estimate. The contribution is the automated linkage of national item coding, time-indexed cost information, ANN forecasting, Revit plug-ins, and BIM quantity aggregation rather than a new optimization algorithm or a probabilistic uncertainty model. The results support the feasibility of the workflow as a proof of concept, subject to stricter chronological validation, improved categorical encoding, benchmark comparisons, and multi-project external validation.