"White Sugar Data for HybridCAE"

"Predicting late white sugar supply is critical for ensuring food security and timely logistics interventions in Indonesia. However, forecasting supply chain disruptions remains challenging because of imbalanced data and complex spatiotemporal dynamics. This study proposes a hybrid forecasting framework that integrates a conditional autoencoder (CAE) with stacked ensemble learning to improve the predictive accuracy for late supply events. CAE captures nonlinear dependencies in 10 years of provincial panel data by learning latent representations guided by feature importance and lagged outcomes. These enhanced features are fused with the original inputs and passed to the XGBoost and CatBoost classifier ensemble. The proposed model achieves superior performance, with an accuracy of 0.948, a precision of 0.893, and competitive recall and F1 scores. The Shapley Additive Explanations analysis confirmed the interpretability of the model, highlighting production and latent features as key predictors. This framework demonstrates the potential of explainable artificial intelligence to support proactive supply chain decisions in agri-food systems."