Demand Prediction in University Booking Systems to Reduce Food Waste Using Neural Networks Including Weighted Error Function

A significant challenge in the university meal booking is the high No-Show rate that leads to considerable food waste in consequence of facing low price of nutrition system and government subsidizing. This study aims to prevent food waste in university dining halls via predicting actual demand.

To model and solve the problem, an Artificial Neural Network has been used that was performed by weighting the error function with Generalized Pattern Search (GPS). Date, the day of the week, the price level of Food, total number of reservations, total number of reservations by undergraduate students, Masters' students, PhD students and dormitory students and the parallel food have been considered as inputs of the model. The output is the actual demands based on Show's number.

The seven-year data of the meal booking system of a large university in Iran has been examined. This data demonstrated that the food waste rate is close to 10% of the total food reservations. An artificial neural network including weighted error function under GPS optimization was obtained to predict actual demand. Finally, the results of training indicated over 80% waste reduction in surplus daily food production.

The proposed model has the potential to provide an estimation of actual demand. Although adding indicators that influence demand estimation, the proposed model is able to change the actual demand prediction at various levels of risk expected by the university. To avoid food waste and prevent the loss of government subsidies, this precautionary approach can control overproduction.