This project features a dataset from kaggle and the analysis done on it to make certain correlations on the type of music and artists with the years and their popularity.
This addresses multiple features of the music that was popular and derives factors such as acousticness, danceability, energy, speechiness, liveness, valence, instrumentalness.
Our analysis shows that audio features do play a vital role in the popularity of a song. A discussion on the trends of music features over the past century has been provided. On testing various models, it is seen that De- cision Tree Regressors, SVM and logistic regression do not yield good results. Random forests and Multi-layer perceptron perform very well even for small-sized samples. With better optimizations and more complex models from deep learning such predictions can be used in the real world to help artists.
The paper in the repository has the entire information on the regression models that we experimented with and the same was presented in a noted conference.