Feedback-enabled digital twin and explainable artificial intelligence framework for predictive resilience and adaptive decision-making in disrupted supply chains
TL;DR
A feedback-enabled digital twin and explainable artificial intelligence framework for predictive resilience and adaptive decision-making in disrupted supply chains demonstrates company-level operational utility for transparent, uncertainty-aware and feedback-enabled supply-chain resilience, while wider multi-company validation remains necessary.
Abstract
Supply-chain disruptions are increasingly difficult to manage with conventional monitor–respond–recover approaches because they provide limited early-warning, explanation, uncertainty assessment and adaptive recovery support. This study developed a feedback-enabled digital twin and explainable artificial intelligence framework for predictive resilience and adaptive decision-making in disrupted supply chains. A quantitative modelling, simulation and multi-objective optimization design was applied to a hybrid dataset containing 3,500 de-identified operational shipment records from CarryGo Logistics Company, Awka, Nigeria, and 1,500 simulation-generated rare-disruption observations. The data were divided into 3,500 training, 750 validation and 750 testing observations, with validation and testing conducted entirely on operational records. Random Forest, XGBoost, LSTM and Bayesian Neural Network models were evaluated, while 10,000-iteration Monte Carlo simulation, SHAP analysis and normalized multi-objective optimization supported uncertainty assessment, explanation and recovery selection. Random Forest achieved 0.79 accuracy, 0.61 precision, 0.99 recall, 0.75 F1 and 0.85 AUC, whereas XGBoost achieved 0.62 precision, 0.86 recall, 0.85 AUC and the lowest MAE of 0.27. The framework classified 353 normal, 61 watch, 333 warning and 3 critical states. Disruption severity increased from 22% at supplier failure to 83% customer-service loss across the simulated network. Supplier reliability and lead time contributed 41.49 and 39.97% of disruption risk, respectively. Inventory redistribution ranked first under equal weights with J = 0.26 and remained preferred under service and risk priorities, while supplier switching became optimal under cost and delay priorities. The framework demonstrates company-level operational utility for transparent, uncertainty-aware and feedback-enabled supply-chain resilience, while wider multi-company validation remains necessary.