<![CDATA[Exports play a critical role in economic growth; however, existing studies often rely on aggregate data and lack integration between predictive analysis and prescriptive decision-making. This study aims to develop an integrated predictive–prescriptive framework for optimizing Indonesian export allocation using hierarchical commodity classification, multi-model forecasting, and linear programming optimization. Using United Nations Commodity Trade Statistics Database (2019-2024), commodities were classified into raw, semi-finished, and finished categories through a Large Language Model-based approach. Forecasting performance was evaluated using SARIMA, Holt-Winters, Random Forest, Gradient Boosting, and XGBoost based on Mean Absolute Percentage Error (MAPE). The results show that model performance varies across commodity stages, where Random Forest achieved 18.11% MAPE for volatile raw shrimp, while SARIMA obtained 8.07% MAPE for stable finished cassava leaves. The forecasting results were integrated into a Gurobi optimization model to generate export allocation strategies. The model increased export destination coverage for eucalyptus leaves from 79 to 130 countries and improved revenue for raw cassava leaves from USD 1,443,290 to USD 1,560,651 despite reduced export volume. This study contributes by explicitly integrating hierarchical commodity classification, multi-model forecasting, and prescriptive optimization into a unified decision-support framework, addressing the limitations of prior studies that primarily focus on forecasting without actionable optimization. However, the model remains sensitive to volatile trade data and does not yet incorporate external factors such as trade policies and regulatory dynamics, which may influence real-world applicability.]]>
