Music success (feat. RDF): A study on music features’ contribution to music success in Spotify through random decision forests.
The growing use of predictive analysis can be seen in volatile industries such as the music industry. Our study contributed to determining which song features are able to assist a song in making it to the top charts with the use of a feature ranking model to establish which specific features are dee...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | text |
| Language: | English |
| Published: |
Animo Repository
2022
|
| Subjects: | |
| Online Access: | https://animorepository.dlsu.edu.ph/etdb_dsi/91 https://animorepository.dlsu.edu.ph/context/etdb_dsi/article/1057/viewcontent/Music_Success__feat._RDF___A_Study_on_Music_Features__Contributio_Redacted2.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | De La Salle University |
| Language: | English |
| Summary: | The growing use of predictive analysis can be seen in volatile industries such as the music industry. Our study contributed to determining which song features are able to assist a song in making it to the top charts with the use of a feature ranking model to establish which specific features are deemed significant. Previous research has only used certain features separately in their respective studies (e.g. main features, auxiliary acoustic features, pitch features, timbre features, or superstar variable). Through this, it was revealed that there are certain features of songs that outperform the other. However, one of the studies added the superstar variable and found noteworthy contributions to their research, generating a higher accuracy rate than other research. We extracted the publicly available data such as the main features, auxiliary acoustic features, and genres of a song from Spotify’s Web API, Spotipy, between the last week of 2016 to November of 2021. Data was then run and tested in a random decision forest model to assess the prediction accuracy rate. The model generated a prediction accuracy of 95.44% with the superstar feature being the most significant among all 51 features, garnering a value of 0.19. The combination of using significant features from previous studies consequently generated a higher accuracy rate especially with the addition of the superstar variable. Our findings provide a set of compositional styles for music artists and labels to make use of when producing a song in order to achieve music success. Meanwhile, the prediction model aids tech practitioners and researchers in other fields to better understand the application of machine learning in the music industry. |
|---|