Comparing Machine Learning Forecasting Models Based on Accuracy and Efficiency for Predicting Demand in a Food and Beverage Company

Accurate demand forecasting significantly benefits the food and beverage industry – minimizing waste, optimizing inventory, and meeting market demands. Traditional forecasting methods fail to capture non-linear relationships and external factors such as holidays, weather conditions, and macroeconomic factors. This study developed a machine learning-based forecasting framework by working on alcoholic beverages as a representative product category given its growing per capita consumer spending and consumption. The research aimed to evaluate the performance of different ML models, in comparison to one another and other traditional forecasting techniques. The models also incorporated feature selection and hyperparameter tuning to optimize predictions and were assessed through different accuracy metrics. Results demonstrated that XGBoost excelled in both accuracy and computational efficiency. Feature selection using correlation enhanced computational efficiency but led to a slight reduction in forecast accuracy. Additionally, hyperparameter tuning methods of Random Search outperformed Grid Search in both accuracy and execution time. Overall, the study recommended adding more factors, leveraging algorithms for other applications, incorporating hyperparameter tuning, and investing in data.