An Explainable Ensemble Deep Learning Framework for Sales Productivity Forecasting in Palm Oil Supply Chains

Abstract

This research presents a comprehensive explainable artificial intelligence (XAI) framework for forecasting sales productivity in palm oil supply chains. The study integrates multiple machine learning models including XGBoost, Random Forest, and Linear Regression with interpretability techniques such as SHAP and LIME to achieve both high predictive accuracy and stakeholder trust. Using 72 months of historical data encompassing 44 variables across supply-demand dynamics, market pricing, regional rainfall patterns, and network performance metrics, the ensemble model achieved R² = 0.87, RMSE = 10.42 units, and MAPE = 17.8%. SHAP analysis revealed that supply-demand ratio, regional rainfall aggregates, and buying price volatility were the most influential predictors. The framework provides actionable insights for supply chain practitioners while maintaining model transparency, addressing the critical gap between predictive accuracy and interpretability in agricultural commodity forecasting.