What Makes a Song Popular: Analysis Using Various Machine Learning Algorithms

Abstract

The music consumption habits are rapidly evolving with the advent of streaming services, for artists, producers, and marketers, the question about what makes a particular song popular becomes invaluable. In this study, the authors aimed to evaluate the popularity of 2,000 songs by means of a machine learning algorithm and were given the Spotify dataset for this purpose. Five techniques – Decision Trees, Random Forests, Linear Regression, XGBoost, and K-Means Clustering - were applied, and prediction power was checked based on accuracy, precision, and recall. The authors successfully identify also the impact of prominent attributes such as tempo, loudness, music style and energy on the level of popularity of the songs in the market, through a combination of feature analysis and model testing. It appears that ensemble models, Random Forest and XGBoost, would be the methods of choice for song popularity prediction because of their high level of performance and versatility to inputs, indicating such co-existing complexity within features of audience appeal.