Three PhD candidates from the University of Trieste, Ruben Viduli, Davide Zennaro and Edoardo Insaghi, have won the “Overall” award at the ninth edition of Stats Under the Stars – SUS 2026, the scientific hackathon promoted as part of the SIS-FENStatS 2026 Joint Meeting, which brought together in Rome the 53rd Scientific Meeting of the Italian Statistical Society and the first Conference of the Federation of European National Statistical Societies.

The competition, held at Sapienza University of Rome with the support of Anas as partner and data provider, challenged young researchers to apply statistical analysis to a real-world problem. Participating teams were given access to a dataset covering reports recorded on the road network between 2016 and 2023, including accidents, potholes, critical issues and other anomalies reported by road users.

The goal was to build a model capable of predicting, for each road and each month, the number of reports expected in 2024 and 2025. Anas already had access to the data for those two years, but these figures were withheld from the participants: once the teams submitted their forecasts, the organisers compared them with the actual values to assess the accuracy of each model.

The UniTS team received the “Overall” award, assigned to the group achieving the best balance between forecasting accuracy and quality of technical analysis. The competition also included awards for the best prediction and the best report.

The challenge took place over the course of one night: participants worked from 7 p.m. until 7 a.m., before returning to Sapienza for the presentation of the results and the award ceremony, held before the official opening of the conference.

From a methodological point of view, the task involved predicting count data, namely the expected number of events. The UniTS team chose a probability distribution suitable for modelling this type of data and introduced an extension to make it more flexible and better able to capture the characteristics of the dataset.

Alongside this approach, the PhD candidates also tested more complex machine learning and deep learning models. During the evaluation phase, however, the simpler statistical model proved to be the most effective. The result highlights how, in some applications, the decisive factor is not the complexity of the algorithm, but the ability to choose the model that best fits the problem.

“It was an intense challenge, completed in just a few hours and based on real data,” the three PhD candidates commented. “The most interesting outcome was seeing how a well-chosen statistical model could compete with much more complex approaches. The differences between the top teams were extremely small, down to the fourth or fifth decimal place, which made the recognition even more meaningful.”

The team brought together skills developed through different academic backgrounds, spanning statistics, mathematics and data science. This combination of perspectives enabled them to integrate methodological insight and applied expertise, producing a forecast that was particularly close to the real data according to the statistical metric defined by the organisers.

Applications of this kind can have very concrete implications for infrastructure management. In the case of the road network, being able to predict in advance where reports are likely to increase could help plan inspections and maintenance more effectively, improving service management and helping to prevent critical situations. More broadly, data-driven predictive models can support strategic decision-making in many other areas, from public services to industrial applications.

The award assigned to the UniTS team has a value of 1,000 euros.