Less Food Waste with more AI at the Bayreuth University Canteen

The "Bayreuth AI Association" is a collective of students at the University of Bayreuth who are passionate about Artificial Intelligence. They deal with Large Language Models, algorithms, machine learning—and questions such as how many portions of potato salad will be served in the canteen next week. They aim to promote sustainability with the help of artificial intelligence. The SWO is on board: "An AI-supported forecast of sales numbers in the canteen would have significant added value for us regarding our planning and the prevention of food waste," says Andreas Voigt, the head of the catering operations at SWO.

Conscious resource management is indispensable for ecological and economic reasons. Optimized inventory management can contribute to such sustainability, as simply less needs to be thrown away. For inventory management in a large kitchen like the canteen at the University of Bayreuth, a precise estimate of the sold meals and thus the consumption of specific foods is crucial. Against this backdrop, Pascal Fechner, a doctoral candidate in the research group for Business and Information Systems Engineering and Value-Oriented Process Management at the University of Bayreuth and one of the founders of the "Bayreuth AI Association," launched the project "Machine Learning 4 Mensa at the University of Bayreuth" (ML4Mensa).

The first planning meeting between members of the "Bayreuth AI Association," Andreas Voigt, and Leon Leichsenring, the head of controlling, purchasing, internal audit, IT at SWO, took place in the fall of 2024. Since then, the members of the student group have already developed a preliminary model that accurately predicts the sale of hot beverages in the canteen: "Each day, we are able to predict the sale of coffee with an average accuracy within 17 cups. The best estimate from historical data is 29 cups on average. With machine learning, we can nearly halve the error and provide a much more accurate estimate," reports Fechner. From this, the AI Association developed a forecast for coffee sales in 2025. It was so promising for cafeteria manager Voigt that he decided to extend the project to meals. "I am convinced that AI solutions will increasingly play a major role in communal catering in the future. I am pleased that there is a group on campus, the Bayreuth AI Association, that is developing such solutions in their spare time and driving technological innovation," says Voigt.

The students are now developing a program into which the names of the regularly recurring meals and the planned sale day are entered. With this and with data from the past - which dish was sold how often on which day in the last 48 months? - the algorithm calculates the probable number of meals that will be sold. "The forecast is based primarily on the historical data of food sales in the canteen since the coronavirus pandemic. Since the dishes repeat, we have several data points for the same meal, from which the algorithm can learn. The program also incorporates other influencing factors such as the season, the weather, the number of students at the University of Bayreuth, and the time in the semester—i.e., whether it is lecture period or lecture-free time," explains Fechner. Results are expected by SWO and AI Association in the fall of 2025.