A Bayesian learning approach for predictive resilience in engineer-to-order supply chains

Accurate supplier lead time prediction is critical for maintaining resilience in Engineer-to-Order (EtO) supply chains, characterized by high customization and uncertainty. This study develops a simulation-based predictive framework combining log-normal sensitivity analysis, Internet of Things (IoT)-driven adaptation, and Bayesian Neural Network (BNN) updating to conceptually investigate predictive resilience. Using industry-informed synthetic data that reflect realistic variability in lead times and operational disruptions, the framework is demonstrated through Monte Carlo simulation conducted across sixteen parameter scenarios under both moderate and high variability conditions, providing a proof-of-concept tool that illustrates potential operational benefits in EtO supply chains and establishes a foundation for future empirical validation. Results show that the baseline log-normal model performs adequately in stable conditions, its accuracy deteriorates under parameter shifts, and the IoT-adjusted framework reduces sensitivity to variability, while the BNN-enhanced model further improves robustness by jointly modeling aleatoric and epistemic uncertainty. The approach advances supply chain analytics by integrating statistical modeling, real-time IoT feedback, and Bayesian learning, offering theoretical insights and simulation-based, conceptual decision-support implications for supplier management and risk analysis. • Develop a resilient lead time prediction framework using analytics for Engineer-to-Order supply chains. • Model supplier lead times with log-normal sensitivity analysis for uncertainty quantification. • Integrate IoT-driven calibration to enhance predictive accuracy and adaptive decision support. • Apply Bayesian learning to capture both aleatoric and epistemic uncertainty in forecasts. • Demonstrate predictive robustness and resilience through Monte Carlo simulation and comparative analysis.