Unveiling Rider Performance Patterns in Motorsport applying Machine Learning techniques to Telemetry Data

Motorsport is one of the most competitive environments, in which small details can have a very significant impact on the final performance. In recent years data started to play a crucial role in almost every domain, and motorsport is no exception; every vehicle nowadays presents a wide variety of sensors to monitor different aspects during the race, with the purpose of enhancing the performance by applying data-driven methodologies at the cutting edge. Starting from telemetry data gathered during official competitions by 2WheelsPoliTO, a racing motorcycle university team from Polytechnic of Turin, we developed a complete pipeline for the analysis of telemetries with the purpose of extracting useful metrics and insights regarding the rider behavior and performance using Machine Learning algorithms, with a major focus on time series segmentation and clustering. Our work is centered in comparing different models already present in literature that are designed for time series processing, discussing their strengths and weaknesses; after carrying out a qualitative evaluation of the output we opted for Toeplitz Inverse Covariance-Based Clustering (TICC), able to provide a very precise description of the track following a completely unsupervised approach. With the obtained segmentation and clustering, we compare the laps and detect differences in terms of actions taken by the rider; this allows us to design custom metrics to evaluate the similarity, consequently understanding what were the inefficiencies and where they are located on the track. The proposed pipeline is intuitive and simple to implement, exploiting a model that is not sensitive to noise and can be applied to different data distributions, preserving the quality of the results. Our proposed approach can be useful in speeding up the process of telemetry analysis, providing insights that can be used to improve the rider actions. Moreover, the information extracted by the pipeline can also be exploited to aggregate different tracks together, thus helping in the decision making process concerning race strategies.