<![CDATA[Increased demand fluctuations pose a major challenge in supply chain management, particularly in the fast-food beverage industry like PT. XYZ. This research aims to build and evaluate a demand forecasting model based on machine learning, considering multivariate variables such as product price, seasonal trends, weather, per capita income, population, and historical sales data. The three algorithms used are Random Forest Regressor, Gradient Boosting Regressor, and Prophet Time Series Model. This research method employs a quantitative approach with descriptive-predictive analysis based on time-series data. Model evaluation was conducted using MAE, MSE, RMSE, and MAPE metrics. The research results indicate that Prophet has the highest accuracy (MAPE: 2.33%) and excels in capturing seasonal trends, while Random Forest ranks second (MAPE: 2.47%) with an advantage in comprehensively handling multivariate variables. Gradient Boosting yields the lowest accuracy (MAPE: 2.70%). The conclusion of this study recommends the use of Prophet for short-term seasonal-based predictions, while Random Forest is more suitable for medium to long-term strategic planning. The combination of the two has the potential to become an accurate and adaptive hybrid approach for optimizing the demand forecasting system at PT. XYZ.]]>
Copyrights © 2026
