Analysis of Spotify's Audio Features Trends using Time Series Decomposition and Vector Autoregressive (VAR) Model

Abstract

Streaming is the most popular music consumption method of the current times. As the biggest streaming platform based on subscriber number, Spotify stores miscellaneous information regarding the music in the platform, including audio features. Spotify’s audio features are descriptions of songs features in form of variables such as danceability, duration, and tempo. These features are accessible via Application Programming Interface (API). On the other hand, Spotify also publishes their own charts consisting of 200 most streamed songs on the platform (based on regions) which are updated daily. By combining Spotify’s song charts and the songs’ respective audio features, this research conducted analysis on musical trends using time series modelling. First, the combined data is decomposed to extract the trend features. Second, a Vector Autoregressive (VAR) model is built and followed by forecasting of the audio features. Lastly, the performance of forecasted values and the actual observations is evaluated. As a result, this research has proven that musical trends can be forecasted in the future for a short period by using VAR model with relatively low error.