Using Soccer Athlete GPS Monitoring Data to Visualize and Predict Features

Abstract

Football is a globally popular sport with millions of players and fans engaging in the game across all levels of competition. As one of the world’s most-watched sports, football demands constant improvements in data analysis tools. This master’s thesis presents a comprehensive pipeline for feature extraction, data visualization, and injury prediction, utilizing GPS data collected from two Norwegian women’s soccer teams. It outlines the development of a systematic process, commencing with preprocessing and feature extraction from raw GPS data, to facilitate subsequent analysis and model training. This process culminates in the creation of two distinct datasets - ’Session’ and ’High Intensity Run’ - which offer invaluable insights into player performance and physical attributes. The study then delves into the creation of several visualization tools, utilizing a mix of the aforementioned datasets, raw data, subjective performance data, and match data. The resulting visualizations serve diverse purposes, providing insights into high-intensity runs, player positions, team heatmaps, and the relationships between subjective game performance, objective GPS metrics, and match data. These tools exhibit potential in assisting players, coaches, medical staff, researchers, and sports scientists in a multitude of scenarios, such as managing tactics, preparing for high-intensity periods, and evaluating player mindsets. Lastly, the thesis explores injury prediction through the deployment of various machine learning models. After testing several models, including Logistic Regression, Decision Tree, xGBoost, LSTM, GRU, and ROCKET, the ROCKET model is found to outperform others for the given dataset, with precision of 0.4167 and recall of 0.4545 (TP:5, TN:2978, FP:6, FN:7). However, the model’s performance is found lacking in consistently predicting injuries, thereby underscoring the need for continued research in this field. This study’s comprehensive process and findings contribute significantly to enhancing our understanding of the application of GPS data in professional sports, while pinpointing areas for future investigation.