Music hit prediction and recommenders project

I authored two Jupyter notebooks for Inspirit AI's AI Scholars program for high schoolers. The two notebooks (numbered 3 and 4) do the following:

1. Notebook 3 takes a dataset of songs on Spotify, labelled according to whether they were a hit or not, and uses the first minute of audio from each song (represented in terms of the 12 timbre components sampled every 500 milliseconds) to predict this label. The notebooks use several supervised learning classifiers, including logistic regression, support vector machine, k-nearest neighbours, Naive Bayes classifier, a multilayer perceptron and a recurrent neural network with LSTM cells.
2. Notebook 4 is a bonus notebook that discusses how Spotify's timbre components were derived, and teaches the students how Principal Components Analysis works.

While I am unable to share the notebooks publicly, please get in touch with me via email if you want to find out more about these notebooks.

I am also working on the problem of hit song prediction and music recommendation beyond the scope of the Inspirit AI curriculum.