Clustering of MotoGP track curves using machine learning techniques

Machine learning techniques are increasingly being applied to new areas. Very recently, the world of motor-sport has also started to approach artificial intelligence. In this context we know how important it is to be able to bring continuous developments in a short time. The performances of the drivers of the various teams are generally very close to each other and be able to make the most of the data collected at each single lap of the track can bring great benefits to others. Using these types of algorithms allows to train models only once and can then be used to predict new data, saving valuable time. In this thesis, a clustering analysis of the curves of twenty MotoGP tracks will be presented. Understand which different circuit curves are similar to each other can be useful, for example, to define the best set-up for the motorbike. The best approach to address this type of work would have been a dynamic one, analysing the telemetry data directly in terms of space-time, i.e. analysing the variation of the data within the track curves. However, the aim of the project proposed by Aprilia Racing srl was to use their statistical indices as features for the analysis. This is because, on the one hand, using telemetry data directly would have taken too long and on the other hand, the creation of these statistical indices had required the effort of various sectors of the company and they wanted to exploit them to get a dataset that could be used for this kind of analyses. This decision led to various problems in the data due to the nature of these indices and the static view of the problem. Nevertheless, after a deep cleaning of the data, we were able to obtain excellent results both in terms of interpretation and generalisation to new curves.