Analysis of Spotify Song Popularity Based on Audio Features Using Random Forest

Abstract

The rapid growth of digital music streaming platforms such as Spotify has significantly increased competition among songs, making popularity an important yet difficult aspect to predict. Understanding the factors that influence song popularity is essential for musicians, producers, and digital platforms in developing effective promotion strategies and recommendation systems. This study aims to analyze the relationship between Spotify audio features and song popularity using a data science approach. The dataset used in this study consists of songs described by various audio features, including danceability, energy, loudness, tempo, acousticness, instrumentalness, valence, and track duration, with popularity serving as the target variable. An exploratory data analysis (EDA) was conducted to examine the distribution of popular and non-popular songs, analyze correlations among audio features, and visualize the relationships between selected audio features and popularity. The results show that the dataset is highly imbalanced, with non-popular songs dominating the overall distribution. Correlation analysis indicates strong relationships between certain audio features, particularly between energy and loudness, while the linear correlation between individual audio features and popularity is relatively weak. Scatter plot visualizations suggest that popular songs tend to have higher levels of danceability, energy, and loudness compared to non-popular songs. However, no single feature can adequately explain popularity on its own, suggesting that a combination of multiple audio characteristics influences song popularity. This research provides an initial insight into the relationship between Spotify audio features and song popularity and serves as a foundation for future studies applying machine learning models, such as Random Forest, for popularity prediction.