Search-guided regression ensembles for accurate, interpretable, and uncertainty-aware construction cost estimation.

Area of Science:

• Construction Management
• Data Science
• Machine Learning

Background:

• Construction cost estimation faces challenges due to complex data and uncertainty.
• Existing methods lack accuracy, interpretability, and robust uncertainty quantification.

Purpose of the Study:

• To propose a novel hybrid framework, Search-Guided Regression Ensemble (SGRE), for accurate and interpretable construction cost estimation.
• To integrate dynamic learner selection, uncertainty quantification, and explainable AI (SHAP).

Main Methods:

• Developed SGRE framework integrating six base learners (KNN, DT, NGB, SVR, MLP, BR).
• Introduced two ensemble variants: Forward Search-Guided Regression Ensemble (F-SGRE) and Backward Elimination Search-Guided Regression Ensemble (BE-SGRE).
• Employed SHAP for model interpretation and prediction intervals for uncertainty quantification.

Main Results:

• SGRE achieved superior prediction performance over traditional single and fixed ensemble models.
• The framework produced well-calibrated prediction intervals, offering reliable uncertainty bounds.
• SHAP analysis identified 'Formwork' as the dominant cost driver, followed by Tributary Area and Concrete.

Conclusions:

• SGRE establishes a robust, explainable, and uncertainty-aware paradigm for construction cost estimation.
• The framework enhances transparency and practical trustworthiness in cost prediction.
• Supports resilient infrastructure, sustainable transportation, and resource efficiency in construction.