Determination of the production amount of perishable product under uncertain demand: Practice in a catering company

In this study, an artificial neural network - based forecasting model was developed to estimate the production quantities of perishable food items included in the monthly menu of a catering company under uncertain demand conditions. The objective was to minimize food waste while effectively meeting customer demand. The model utilized time variables, food types, and pairwise interaction variables as inputs, aiming to separately predict the daily demand for five food categories: soup, main course, side dish, dessert/fruit, and salad/yogurt. The test results, with R² 0.94 and MSE 18.98, demonstrate that the model achieved high prediction accuracy. The developed system is capable of operating under dynamic menus composed of different food types each day. Its ability to produce limited predictions even for food types not included in the training data indicates a significant advantage over heuristic approaches. The model contributes to reducing food waste by aligning production more closely with actual demand and performs reasonably well in minimizing unmet demand. To further enhance model performance, it is recommended to include triple interaction variables with the integration of external factors; additionally, methods such as data augmentation, embedding vectors, and weight transfer can be employed to address unlearned components. Keywords: Demand Forecasting, Artificial Neural Network, Perishable Food Products, Interaction Variables, Menu Planning