Abstract:

The aim of this study was to collect training data from e-cycling athletes and conduct a machine learning process to predict an athlete's level of preparation for specific competitions, taking into account their training cycle. The study collected and analyzed data from a three-month training period. Experiments were conducted using three supervised machine learning methods, which achieved an average classification accuracy of 0.6. The results indicated that the developed dataset allows for satisfactory prediction of athletes' level of preparation for competitions. Additionally, an unsupervised learning model was used to identify clusters of training cycles in the feature space, which allowed for differentiation between groups of athletes based on their preparation. The results suggest that achieving higher model accuracy requires collecting more extensive data covering a broader group of athletes and additional characteristics, including external factors such as well-being, nutrition, and environmental conditions. Taking these variables into account could significantly improve the quality of predictions obtained using machine learning models.