Lessons learned from organizing an in-classroom forecasting challenge for teaching uncertainty quantification

In an effort to foster interdisciplinary and innovative teaching via a ‘gamification’ approach, a real-time forecasting challenge was organized during the winter term 2021/2022 at the Karlsruhe Institute of Technology. The main aim was to teach students about statistical methods, probabilistic forecasting and uncertainty quantification in a practical, real-world problem-oriented setup. Around 20 participants (mostly MSc-level students) from backgrounds in mathematics, economics, computer science and geosciences were tasked to provide probabilistic predictions of several targets, including temperature and wind speed at a local weather station in Karlsruhe, over 14 weeks during the semester. Real-time feedback was provided in the form of automated evaluation and rankings of the participants’ submissions, who were competing for a ‘Student Award’ sponsored by the International Institute of Forecasters. Efforts and problems in building and evaluating models were discussed in weekly virtual meetings.

In this presentation, I will discuss the setup and relevant design choices of the forecasting challenge, along with the main outcomes and lessons learned from the practical implementation in the context of the course. All relevant code and material are available as open educational resources.