This independent research project is the first of its kind to experiment with predicting hit songs with a multi-input neural network. To do this, I crafted the most extensive multi-modal dataset for this task, combining both audio and lyric features.
Data collection was distributed among multiple EC2 instances on AWS and the resulting data was stored in a Postgres database with RDS.
The experiments and their results are in Spotify_Analysis.ipynb.
The resulting accuracy of the neural network was 86.4%, which is the highest accuracy of any NN in the hit prediction domain.