Integrated BIM-Based Cost Forecasting for Construction Projects Using AI

Thesis Abstract

Integrated BIM-Based Cost Forecasting for Construction Projects Using AI This study investigates the integration of Building Information Modeling (BIM) with artificial intelligence (AI) to enhance cost forecasting accuracy and timeliness in construction projects. The problem addressed is the persistent discrepancy between initial cost estimates and actual expenditures, driven by fragmented data, limited historical insights, and dynamic project conditions. The aim is to develop an AI-enabled BIM cost-forecasting framework that leverages integrated project data to produce progressive, probabilistic cost estimates throughout the project lifecycle. Specific objectives include (i) identifying data requirements and interoperability standards to enable BIM-AI integration, (ii) developing a hybrid forecasting model that combines machine learning techniques with parametric cost drivers, (iii) validating the model using multi-project historical datasets and prospective case studies, (iv) evaluating the model’s performance against conventional estimation methods, and (v) delivering practical guidance for industry adoption, including governance and risk considerations. A sequential explanatory mixed-methods approach is employed. The population comprises 60 completed and ongoing construction projects across commercial and infrastructure sectors from two large global contractors. A purposive sample of 40 projects with rich BIM datasets and cost histories is analyzed, while 20 projects are used for prospective validation. Data collection integrates BIM models, cost records, schedule data, material quantities, and change orders sourced from the client’s information management system and the contractor’s ERP, supplemented by interviews with 12 project controllers to capture tacit cost drivers. The research design combines quantitative model development and qualitative validation. The quantitative phase applies feature engineering to extract cost drivers from BIM (e.g., quantity take-offs, assembly-level costs, reinforcement schedules), followed by the development of a hybrid forecasting model that fuses gradient boosting regression, Gaussian process regression for uncertainty quantification, and a Bayesian updating mechanism within a CPT-based probabilistic framework. The qualitative phase employs thematic analysis of interview transcripts to identify non-quantifiable drivers, governance practices, and data quality issues, informing model refinement. Validity and reliability are ensured through cross-validation, out-of-sample testing, and robustness checks against alternative model specifications. The analysis employs regression diagnostics, mean absolute percentage error (MAPE), root mean squared error (RMSE), and Brier scores for probability estimates, complemented by scenario simulation to assess forecast sensitivity under schedule and material-price volatility. The study also applies the Theory of Planned Behavior and Resource-Based View to interpret adoption willingness and data capabilities as antecedents to effective BIM-AI integration. A conceptual model maps data flows between BIM, cost databases, and AI components, with feedback loops for continuous learning and update. Expected findings indicate that the integrated BIM-AI framework reduces cost forecast error by 28–35% and decreases forecast lead time by 22–30% compared with traditional 2D estimation and baseline BIM methods. The model is anticipated to provide probabilistic cost trajectories with controllable uncertainty bands (95% credible intervals) and real-time alerts for forecast deviations triggered by key drivers such as design changes, procurement lead times, and price volatility. The research is expected to reveal critical data quality factors, interoperability challenges, and governance requirements essential for reliable forecasting, including standardized data schemas, model explainability, and clear ownership of BIM data segments. Contributions to knowledge include (i) a novel hybrid BIM-AI cost forecasting model that integrates machine learning with probabilistic risk assessment within a BIM-enabled workflow, (ii) an empirically validated methodology for transforming BIM data into actionable cost insights across project phases, and (iii) practical governance and data management guidelines to support scalable adoption in industry. The study closes with recommendations for practice, including standardized data governance, BIM execution plans aligned with AI objectives, and a stepwise implementation roadmap emphasizing data quality, stakeholder engagement, and continuous model monitoring. The findings support policy and practice by demonstrating how AI-augmented BIM can transform cost management from a reactive to a proactive discipline, enabling more accurate budgeting, improved change control, and enhanced project performance.