AI-powered BIM for Optimized Cost Estimation and Risk Analysis

Thesis Abstract

This study addresses the persistent gap between BIM-enabled design data and accurate, actionable cost estimation and risk assessment in construction projects, where traditional methods fail to integrate multidisciplinary inputs and dynamic project changes. The aim is to develop an AI-powered BIM framework that automatically generates optimized cost estimates and risk profiles by fusing geometric BIM data with historical cost databases, schedule information, and risk event ontologies. Specific objectives include (1) to design a machine learning–driven cost estimation module that leverages gradient boosting and deep neural networks to predict direct and indirect costs from BIM-derived quantities; (2) to implement a probabilistic risk analysis component that quantifies cost and schedule risks using Bayesian networks and Monte Carlo simulation; (3) to develop an adaptive feedback mechanism that updates estimates as project data evolves through the lifecycle; (4) to evaluate the integrated framework against conventional QS practices using real project case studies; and (5) to validate the framework against established decision-support criteria and theoretical constructs from contingency theory and information processing theory. The methodology adopts a mixed-methods approach within a multiple-case study design. The population comprises 12 completed and ongoing medium-to-large-scale construction projects across commercial, civil, and industrial sectors. A purposive sample of 6 projects with rich BIM datasets, cost histories, and risk incident logs will be analyzed, supplemented by 6 project managers and quantity surveyors for expert validation. Data collection instruments include (a) BIM extracts and project cost databases, (b) a structured survey instrument for eliciting expert judgments on risk factors, and (c) semi-structured interviews to capture procedural insights and contextual factors. Instrument validity will be established through content validity by a panel of five senior QS academicians and practitioners, and reliability will be assessed via test-retest procedures and Cronbach’s alpha for survey items. Analytical techniques comprise (i) machine learning models—gradient boosting machines (XGBoost) and multi-layer perceptron networks—to estimate cost components from BIM-derived quantities and attributes; (ii) Bayesian networks to model probabilistic causal relationships among risk drivers and their impact on cost and schedule; (iii) Monte Carlo simulation to propagate uncertainty through the cost model; (iv) regression analysis to benchmark performance against baseline QS methods; and (v) thematic analysis of interview data to extract organizational and process-related determinants. The conceptual model integrates resources, IT infrastructure, and decision-making processes using Resource-Based View theory and information processing theory to explain how AI-enhanced BIM capabilities influence cost accuracy and risk visibility. The study will apply cross-validation, SHAP (SHapley Additive exPlanations) values for feature importance, and out-of-sample testing to ensure generalizability. Ethical considerations include data anonymization, informed consent from participating professionals, and adherence to data governance and confidentiality standards. Expected findings indicate that the AI-powered BIM framework reduces mean absolute percentage error (MAPE) in cost estimation by 18–26% relative to traditional QS methods, enhances the precision of contingency allocations by enabling real-time risk updates, and improves project decision speed by 22–35% through integrated visualization dashboards. The Bayesian network is anticipated to identify the most influential risk drivers (e.g., design late changes, supplier lead times, and site conditions) and quantify their contributions to cost overruns, while Monte Carlo simulations provide probability distributions for budget-at-completion scenarios under varying risk appetites. The research is expected to demonstrate that AI-augmented BIM fosters better alignment between design intent and cost outcomes, supports proactive risk mitigation, and strengthens data-driven governance across stakeholders. The study contributes to knowledge by bridging AI, BIM, and QS practice, extending the application of Bayesian risk modeling in construction cost management, and validating a scalable, lifecycle-oriented framework that can be integrated with existing ERP and BIM platforms. Conclusions point to the necessity of robust data governance, model interpretability, and ongoing professional development to maximize benefits. Recommendations include developing industry standards for BIM-embedded cost and risk data exchange, investing in interoperable data schemas, and piloting the framework on larger-scale projects with diverse procurement routes to further test scalability and resilience.