<![CDATA[Accurate demand forecasting is a crucial factor in culinary industry supply chain management. This study compares the performance of two ensemble learning methods, namely Random Forest and Gradient Boosting, in predicting demand for six culinary products with different demand characteristics. The data used consists of monthly sales data at CV Healfit Pangan Sehat (DietGo Kitchen). Models were evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). The results show that Random Forest delivered superior performance for 5 out of 6 products with an average MAPE of 31.61%, compared to Gradient Boosting with an average MAPE of 35.43%. Random Forest proved more robust in handling products with stable demand patterns (Sei Ayam: MAPE 19.76%, Sei Sapi: MAPE 21.38%) and intermittent demand (Sei Domba: MAPE 26.88%). Feature importance analysis revealed that lag-3, lag-6, and trend were the strongest predictors in both models. Gradient Boosting outperformed Random Forest on only one product (Sambal Bawang: MAPE 37.16%). High-volatility products such as Baked Grill Chicken yielded a MAPE of 32.98%. This study provides a practical contribution in the form of a forecasting method selection framework based on product demand characteristics, along with recommendations for implementation in the culinary industry.]]>
Copyrights © 2026
