modeling-analysing-songs

This is a project aimed at providing a tool for predicting how successful a song will be based on its audio features, and other engineered features we will be looking at and adding to the models. The first part of the project looks at doing data cleaning on the dataset that we have available, and then we look at feature engineering: song similarity to others, the geospatial aspect of the songs, visualizing the audio features and how they change with different groups of songs. Then, we turn to building models, starting with preparing the data for the linear model, after which we fit and refine a non-parametric model, based on the r2 and mean squared error metrics of how they are performing.