Popularity Prediction of Music Based on Factor Extraction and Model Blending
Yutong Ge, Jiaqian Wu, Yutong Sun
Abstract
The purpose of this research article is to explore different factors that can affect the popularity of a song and to predict the potential popularity of one song with the most related factors. Since there are too many factors that can affect a song's popularity and high dimensions caused by large amounts of variables, they may result in significant errors in analysis. To avoid these errors, the methods used in this research are Principal components analysis (PCA) and model blending. By using PCA, we can get lower dimensions; and then model blending to model the relationship between a scalar response and one or more explanatory variables. Different weights to random forest, decision tree and knn are set to get a new model, with mean squared error (MSE) 4.96, which is a comparatively low one. The results provide a relatively precise prediction of the popularity of Spotify top music. And they can be used in music company's commercial analysis and enrich the field of popularity prediction of the song market.