Projects
ML4Mensa
We are collaborating with the Student Services Oberfranken (SWO) to develop a program that predicts the number of meals sold each day at the University of Bayreuth's canteen. For this purpose, we processed the data from transactions dating back to 2015, and used machine learning to predict the future demand.
By combining feature engineering techniques and advanced foundational models for tabular data, we were able to accurately (R2 = 0,94) predict the demand, even months in advance.
Media coverage:
- Bayreuther Tagblatt (in German)
- Kurier (in German)
- inFranken.de (in German)
- UBT press release in English and in German.
Contributors:
- Pascal Fechner
- Nico Höllerich
- Renato Mio
- Felicitas Feick
- Pascal Lange
This project is in progress.
Coffee consumption forecasting
To support management decision-making regarding hot beverage sales in the University of Bayreuth cafeteria, we predict hot beverage sales for given dates and make a forecast for sales in 2025. As a dataset, we used the sales of hot beverages in the cafeteria of the University of Bayreuth between January 2017 and October 2024, including the sales of coffee, milk coffee, cocoa, and tea.
For our prediction model, we chose an LSTM architecture and a supervised learning approach. While a benchmark model in the form of an educated guess based on past sales achieved a mean absolute error of 28.47 coffee units sold (per day), our best-performing model achieved a mean absolute error of 17.25 units sold.
Contributors:
- Ivan Khrop
- Pascal Fechner
- Andreas Karasenko
- Felicitas Feick
- Renato Mio
- Gianluca Pani Casanova
- Franz Erhard
- Henri Berntgen
- Frederick Hornung
- Nico Höllerich
- Tom Zeilmann
This project was concluded and is currently closed.