Igor Sikorsky Kyiv Polytechnic Institute hosted the “Plasma Code” ML Hackathon, organised by the Faculty of Physics and Mathematics in collaboration with the Abitly platform. The event brought together more than 50 participants in 10 teams, who spent three days applying machine learning techniques to predict plasma disruptions in fusion devices.

Representatives of the Applied Mathematics Department — Iryna Nalyvaiko, Matvii Tereshchenko, Sofiia Nikolaienko, and Ruslan Korytich — were members of a team that secured third place.

The participants analysed data from the European JET tokamak and developed models for the early detection of plasma instability. Addressing this challenge required combining expertise in applied mathematics, data analysis, machine learning, and physics. The goal was not only to identify potentially dangerous changes but also to provide sufficient warning time for an effective response.

The team developed a solution based on tree-based models using specially engineered physics-informed features, supplemented by a separate physics-based component. During optimisation, the participants considered the trade-off between timely disruption detection and false alarms, taking into account their financial and temporal costs.

According to the hackathon evaluation results, the solution achieved a disruption detection recall of 96.4% at a false positive rate (FPR) of 25%, detecting disruptions 300 ms before their occurrence. The developed solution earned the team a prize-winning position and demonstrated the potential of mathematical methods for addressing complex engineering challenges.

We warmly congratulate our participants on this achievement! We wish them continued professional success, exciting research projects, and many more victories!

Read more about the event in the KPI Media report.