The recorded music industry makes billions of dollars globally each year. Streaming services such as Spotify, Apple Music and Pandora now make up a large part of that revenue. Their prot is determined by how many times a song is streamed. Thus, artists, recording labels and streaming services all have an interest in being able to predict which songs will be most popular to maximize their prots. We employed a number of supervised learning methods to predict the popularity of a song based on musical features scraped from Spotify. We focused on the ability of our classiers to correctly classify the top 25% of songs in our data set. Starting with a single decision tree as a baseline with 47.72% test accuracy, our best model was Adaboost with 71.89% test accuracy in classifying the top quartile of songs. The predictions of these models could be used by members of the music industry to make investments in certain songs and artists over others.
NOTE: This was a group project. I only uploaded the share of code here that I personally programmed. The final project report was a team effort, too.