Answers are grounded in this paper and cite their sources.
Purpose The convergence of building information modelling (BIM), artificial intelligence (AI) and Industry 5.0 has increased the need for transparent, resilience-oriented decision support in construction. However, existing BIM–AI studies provide limited interpretability and rarely integrate sustainability, resilience and governance. This study develops a multi-layer explainable AI framework for operational sustainability resilience intelligence in BIM-oriented construction management. Design/methodology/approach The framework integrates sustainability interaction engineering, exploratory dimensional assessment, hierarchical operational sustainability resilience index (OSRI) modelling, explainable ensemble learning and governance-oriented zoning. It was evaluated using a publicly available BIM–AI benchmark dataset covering operational, structural, environmental, resource and safety indicators. Random Forest, XGBoost and LightGBM were evaluated using stratified cross-validation and SHAP-based explainability, alongside leakage-aware sensitivity analyses, alternative OSRI weighting schemes, multicollinearity diagnostics and input-noise robustness tests. Findings LightGBM achieved 89.0% test accuracy, a weighted F1-score of 0.8957 and an ROC-AUC of 0.9684 using the original 15 indicators. However, the test weighted F1-score declined to 0.5062 under a conservative eight-indicator specification excluding potentially target-related predictors, demonstrating substantial sensitivity to predictor selection. Therefore, the results constitute a methodological demonstration rather than independently validated operational risk prediction. OSRI-based zoning identified four conceptual governance categories: resilient, stable, vulnerable and critical. Originality/value The study integrates explainable AI, interaction-aware sustainability modelling, hierarchical resilience assessment and exploratory governance zoning within a unified analytical framework. It provides a reproducible methodological foundation for sustainability-resilience assessment and future validation using independently sourced industrial BIM data and digital twin-enabled infrastructure systems.
No takes yet. Share an insight, caveat, or question.
Admane et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: