EXPLAINABLE ENSEMBLE LEARNING FOR PREDICTING CAPITAL RESILIENCE UNDER SUPPLY CHAIN DISRUPTIONS IN NIGERIAN FOOD AND CONSUMER GOODS SECTOR

This study develops an explainable ensemble learning framework for predicting capital resilience from supply chain disruption dimensions (supply chain orientation, risk management, and resource configuration capabilities) in Nigeria's food and consumer goods sector, and for guiding resilience-related investment decisions. The study draws on survey data from 390 employees across production, procurement, logistics, marketing, and warehousing departments at Okomu Oil Plc, Presco Plc, and Guinness Nigeria Plc in Benin City. To extend the analysis beyond the limits of the cross-sectional dataset, were obtained from the empirical distribution, covariance structure, and implementation patterns observed in the original data. Three ensemble learning models (Random Forest, XGBoost, LightGBM) were implemented and compared with logistic regression and linear SEM. Model performance was evaluated using five-fold cross-validation with AUC, F1-score, and accuracy. SHAP analysis was conducted for model interpretability. The results show that XGBoost performed best, achieving AUC = 0.892 and explaining 74.3% of the variation in capital resilience (dichotomised), compared with 68.4% explained by the original SEM model. The findings also reveal important nonlinear patterns. Risk management showed diminishing returns beyond a threshold of 4.1, supply chain orientation produced stronger benefits when combined with high risk management scores, and resource configuration capabilities showed no significant direct effects but moderated the risk management–capital resilience relationship when RCO exceeded 3.8. SHAP analysis identified risk management as the most important predictor, contributing 47% of predictive power. To make the findings more useful for decision-making, the study developed an investment simulation framework for testing portfolio allocation, sequencing strategies, and breakeven periods for resilience-building initiatives.