Music Genre Classification - Spotify Audio Properties

Summary

The goal of this study is to understand the various Spotify audio features that strongly impact and determine the genre of a song, as well as to build a parametric and non parametric model that can accurately predict the genre of a song based on these audio features. Multinomial Logistic Regression (MLR) and Gradient Boosting Method (GBM) algorithms have been used and the GBM model performs better, having an overall accuracy of 54.02%. Audio features such as Danceability, Acousticness, and Speechness have the highest relative influence when predicting the genre of a song. I also built a R Shiny application where an artist can input their song's audio features and the application will predict the genre of the song. The application will also display the top 10 most similar songs to the inputted song features.