Demystifying the Popularity of Songs Using Machine
Learning Algorithms
Albert Wang
-
Miramonte High School, California, USA
Issue:
Vol. 1 No. 1 (2023)
Date Published:
11-07-2023
Keywords:
Machine Learning, Song Popularity Prediction
ABSTRACT
Song popularity is an influential subject within the modern music streaming industry. It determines which artists can
gain media attraction, gather loyal fans, and ultimately succeed. Analyzing song popularity with ML algorithms
contributes to demystifying success within the music industry. Two datasets, datasets 1 and 2, collected from the
Spotify Web API contain audio information on respectively 2000 songs and 240,057 songs. Ordinary Least Squares
Linear Regression (OLS LR) and Neural Network (NN) algorithms were used on each dataset to predict song
popularity. The most complex NN structure used in this study contains three hidden layers, achieving the best
regression performances on both datasets; however, it was superior to other models by a small margin. Overall,
models trained with dataset 2 achieved superior results, particularly in the R^2 metrics, but were unimpressive due
to low regression metrics.