Integrating AI with BIM for enhanced construction cost and schedule forecasting: a machine learning model with barrier analysis
Asmaa A. Soliman, Ahmed H. Ibrahim, Ali Saeed Almuflih, Boshra TahaPurpose
This research identifies the barriers in BIM-AI adoption and demonstrates how the combination of building information modeling (BIM) and artificial intelligence (AI) can be used to improve construction project schedule and cost forecasting through an integrated framework.
Design/methodology/approach
This study was conducted using a combination of different research methods. The questionnaire survey measured the 43 barriers by means of the relative importance index (RII) and Pareto analysis. A Python-based BIM-AI forecasting framework was developed based on the findings. Random forest, gradient boosting and XGBoost were benchmarked using 5-fold group cross-validation (GroupKFold) with project-level grouping to prevent data leakage.
Findings
The results indicate that random forest and gradient boosting showed more stable performance in duration prediction (R2 = 0.806), while random forest achieved the highest accuracy in cost forecasting (R2 = 0.909) and was therefore selected for integration into the final BIM–AI framework. In the applied case study, the model achieved a duration R2 of 0.7886 and a cost R2 of 0.4368, reflecting forecast variance due to project uncertainties rather than realized cost savings. The barrier adjustment factor was 1.0857 (8.57%), consistent with FIDIC Sub-Clause 13.8, which limits contractual adjustments to 10%.
Research limitations/implications
The BIM–AI model was developed using a dataset comprising 15 construction projects and 845 activities. A total of 8 projects were used for model training and GroupKFold cross-validation, while 7 unseen projects were reserved for independent testing. For presentation purposes, detailed forecasting outputs are illustrated using one representative project (Project A). Forecast deviations (5.88% in cost, 14.3% in duration) represent prediction variance rather than realized savings. Model performance remains dependent on the quality of historical data, and real-time risks, such as price fluctuations, labor changes or supply-chain disruptions, were not considered. This study presents a proof-of-concept BIM–AI forecasting framework validated on historical project data. The model has not yet been deployed in a live project decision-making setting, nor have practitioners used the workflow in real-time. The case study (Project A) serves as an example, showcasing the application of the model on a finished project instead of an assessment of a potential deployment. Future work will focus on live implementation, user feedback collection and integration with existing project management workflows to assess practical benefits and usability.
Originality/value
The study combines RII-based barrier analysis with a machine learning forecasting framework in a BIM environment, validated using project-level GroupKFold.