Uncertainty aware and explainable construction cost prediction using a hybrid probabilistic learning model.

Area of Science:

• Construction Management
• Data Science
• Machine Learning

Background:

• Accurate construction cost forecasting is critical for project success.
• Existing models often lack reliable uncertainty estimation and interpretability.
• Complex feature interactions in sustainable building pose challenges for cost modeling.

Purpose of the Study:

• To present a unified probabilistic framework for construction cost forecasting.
• To address gaps in reliable interval calibration, model interpretability, and robustness.
• To improve decision-making in construction budgeting and risk management.

Main Methods:

• Developed NGBoost-ETR (Natural Gradient Boosting with Extra Trees base learners).
• Trained the model on a real-world RSMeans dataset (4477 samples).
• Validated probabilistic calibration using six quantitative metrics (PICP, PINAW, MPIW, CWC, NLL, CRPS).

Main Results:

• NGBoost-ETR achieved superior predictive performance (R²=0.9866) compared to baselines.
• Demonstrated robust probabilistic calibration and interval efficiency.
• Showcased SHAP-based interpretability for transparent decision-making.

Conclusions:

• NGBoost-ETR offers a validated, process-innovative framework for construction cost forecasting.
• The model enhances resource efficiency and supports risk-aware decision-making.
• This approach promotes transparency in construction budgeting and tendering.