Optimal Machine Learning Algorithms for Predicting the Popularity of Songs

Abstract

Predicting song popularity has become a hot topic of research in recent years due to its necessity. This study investigates the effectiveness of different machine learning models in predicting the popularity of songs on Spotify using audio features. By comparing Linear Regression, Random Forest, and K-Nearest Neighbors (KNN), the research aims to identify the most suitable algorithm for this task. A dataset containing key musical attributes such as danceability, loudness, energy, tempo, and valence was preprocessed through standardization and one-hot encoding. Model performance was evaluated using metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared (R²). The results show that Random Forest outperforms the other models with the lowest prediction error and highest explanatory power. Additionally, feature importance analysis revealed that duration, speechiness, and emotional characteristics like energy and valence are more decisive in determining a song’s popularity, whereas musical key and mode are less influential. The study concludes that while audio features offer valuable insights, external factors such as playlist placement and social media trends should be considered in future work to improve prediction accuracy.