Automated music genre prediction has many interesting uses, including the music recommendation algorithms popular in modern day music applications such as iTunes' Genius Playlists and Spotify's Suggested Tracks. The majority of the algorithms that form the basis of these services are privately held and used by the companies who created them.
We know many of these algorithms use sentimental analysis on the lyrics of songs for their recommendations, but that comes with its own costs. For one, music recommendation companies must consider the storage constraints they may encounter when storing lyrical text for the massive collection of the songs they may recommend.
We hope to find a less data-intensive way of classifying songs into genres (the first and most simple step in grouping songs by similarity) by using song metadata that numerically encapsulates subjective attributes such as energy and danceability. We hope that in using these subjective attributes, we will be able to recover some quality of music classification that is lost when an important portion of the data set is removed: the lyrics.