Predicting Hit Songs with Machine Learning
Minna Reiman, Philippa Örnell · KTH Publication Database DiVA (KTH Royal Institute of Technology) · 2018
Exploring the possibility of predicting hit songs is both interesting from a scientific point of view and something that could be beneficial to the music industry. In this research we raise the question if it is possible to classify a music track as a hit or a non-hit based on its audio features. We investigated which machine learning algorithms could be suited for a task like this. Four different models were built using various algorithms such as Support Vector Machine and Gaussian Naive Bayes. The obtained results do not indicate that it is possible to predict hit songs on our particular dataset. This stands in contrast to some previous research within this field. We discuss the potential problem in using only audio features, and how this seems not to be sufficient information for predicting a hit.