Music is an important component of our daily lives. We dance, sing, enjoy, and cry simply because of music. But what music is right for any given moment? Spotify tries to use song and playlist data to predict this.
In this paper, I provide a preliminary model for predicting the next best song for a music playlist on Spotify.
Specifically, I apply a blend of supervised and unsupervised machine learning methods (including K-nearest-neighbours, neural networks, multi-logit regression, Bayesian multi-logit regression, and gaussian mixture models) to find songs that are most similar to a playlist’s features. The advantages and disadvantages of my approach are discussed, and my final results are displayed.
In conclusion, I realise that there are a few fundamental flaws with my approach, however, the results are somewhat promising and provide unique insight into a variety of machine learning and Bayesian methods for predictive inference.
See the PDF for the details of my research, and the .ipynb file for my code.