Model strojového učení pro predikci úspěšnosti hudebních skladeb

This bachelor's thesis deals with the design and evaluation of a machine learning model for predicting the success of songs on Spotify. The main goal was to create a model capable of predicting a song's success prior to its actual release, with the motivation of supporting the decision-making processes of artists and employees of the music industry regarding music release and promotion. For this purpose, a custom dataset was created, combining parameters from Spotify API, Genius API, MusiCNN and the Essentia library. Neural network, beta regression, and XGBoost models were trained on this data. Emphasis was also placed on interpreting the results and determining the key parameters that define a popular song. The results show that the XGBoost model achieved the highest accuracy. Finally, this model was tested on real data.