Analyzed a dataset that I created, comprising of different attributes of Prince's music through the eighties - song length , BPM , Key etc.
This project was in collaboration with Jamie Wheeler - http://www.wheelerdatalab.com/ .
Process
Wanted to understand if there was a pattern to the music that an artist composed through which we could predict if the song that was being written was going to be a hit or not. The min reason we chose Prince was because his songs are very diverse.
We considered the target variable to be if the song was in the Hot 100 billboard list or not, while the categorical variables were the attributes of his song , like BPM ( Beats per minute ) , song length , Key of the song etc .
Performed linear and logistic regressions on the data . Found a few correlations with song key and song structure , but no correlation with the song being a hit or not.
Also plotted a decision tree to understand if there was a way we could classify the kind of music he made.
Major findings
There are relationships in album structure or song structure, but nothing that can predict if his song would be a hit.
There is some structure to his albums as regards song length with a preference to increase song length as the album progresses.
Tracks 4 and 9 are his preferred neutral sentiment track. Track 9 on Purple Rain, “Purple Rain” is a perfectly neutral sentiment song.
Not a lot of descriptive steps necessary to classify his songs as a hit categorization.
If you're interested in reading the longer version : https://github.com/saurabh-rao/Computer-Blue---Analyzing-Prince-s-music-through-the-eighties-/tree/master/Report