A subset of the Million Song Dataset is used. The dataset available on Kaggle.
An initial trend analysis of the Billboard charts from 1950 to 2015 : Evolution of pop music
Prediction of the release year of a song given its timbre averages and covariances, using linear regression. Release Year Prediction on MSD
We try to reduce the prediction error by implementing a neural network. Release Year Prediction on MSD (using neural nets)
These were trained using Kaggle kernels and the links to those are available here and here respectively.