Voors, Jan (2026) Using audio features in a deep learning model to predict music popularity. Bachelor thesis, Data Science and Society (DSS).
|
PDF
BA5660645JWVoors.pdf Download (1MB) | Preview |
Abstract
Music popularity prediction has been a point of interest for researchers and artists alike. In this research paper, the predictive capabilities of a Convolutional Neural Network were combined with the Free Music Archive in an attempt to research how successfully music popularity can be predicted using deep learning models. Different features were extracted from the audio dataset, like MFCC features, spectral contrast, and Chroma features. Different iterations of the model were created, and both a 1-dimensional and 2-dimensional version were developed, to ensure the maximum potential of the proposed model. Its performance is compared to basic machine learning architectures, and put up against a more well-tuned Random Forest model. Ultimately, the performance of the proposed model proved to be too insignificant to display any meaningful predictive correlation, displaying limitations in the data’s capabilities.
| Item Type: | Thesis (Bachelor) |
|---|---|
| Name supervisor: | Do, T.P. |
| Date Deposited: | 10 Jun 2026 11:44 |
| Last Modified: | 10 Jun 2026 11:44 |
| URI: | https://campus-fryslan.studenttheses.ub.rug.nl/id/eprint/814 |
Actions (login required)
 |
View Item |