Project Cost Estimation Based on BIM and Deep Learning

Abstract The accuracy of project cost estimation directly affects the effectiveness of construction management, and the nonlinear fluctuation of material prices is a major source of estimation bias. To address this issue, this paper proposes a dynamic cost estimation framework that couples 5D quantities extracted from building information modeling (BIM) with a grey BP neural network optimised by the sparrow search algorithm (SSA‑PGNN) for material price forecasting. Unlike traditional static approaches, the framework introduces a bidirectional event‑driven data interface and a dual‑threshold feedback mechanism that support dynamic updating of cost estimates and proactive risk immunisation. Using data from 62 real construction projects in Liaoning Province, China (2016–2021), material quantities are extracted from BIM models and monthly material prices are forecast by the SSA‑PGNN model. The proposed model achieves a mean absolute percentage error (MAPE) of 2.43%–4.04% for material prices and 2.9% for total project cost under leave‑one‑project‑out cross‑validation, outperforming the PGNN, SSA‑BP, XGBoost, random forest and LSTM benchmarks (p < 0.05) within the present dataset. Time‑series rolling validation, a synthetic extreme‑shock stress test and statistical hypothesis testing support the stability and robustness of the proposed method; however, as the evidence derives from a single regional dataset, external validation beyond the study region remains an essential next step.