How to Make a Hit on Spotify: Insights from Data Analysis

In today’s digital music landscape, understanding what makes a song popular on streaming platforms like Spotify is crucial for artists, producers, and record labels. Our analysis of 170,000 songs released between 1921 and 2020 uncovers the key factors that drive success on Spotify. 

What Defines Popularity on Spotify?

Spotify's popularity index (0-100) measures a song’s success based on several factors, calculating based on:

• Streams
• Engagement
• Recency
• Playlist Inclusion

For our study, we defined a hit song as having a popularity score of 50 or higher—roughly the top 25% of all songs in the dataset.

The Evolution of Popularity Over Time

One key trend in our analysis was the steady rise in song popularity over recent years. This can be attributed to Spotify’s launch in the U.S. in 2011, which transformed the way people consume music. However, older songs tend to have lower popularity scores, suggesting that recency is a significant driver of success.

Which Features Matter Most for Success?

Our study examined 10 key audio features, such as energy, danceability, loudness, speechiness, and valence, to determine their impact on popularity.

• Energy: Higher-energy songs tend to perform better.
• Loudness: Loudness correlates strongly with energy, making it an important predictor.
• Danceability: While not as influential as energy, it still plays a role.
• Acousticness and Instrumentalness: Popular songs are more likely to have higher energy and fewer acoustic elements.

Building a Predictive Model for Popularity

To predict a song’s popularity, we tested several machine learning models:

Linear Regression (R² = 0.759, MAE = 7.98)

MODEL 1: PREDICTING POPULARITY BASED ON SONG FEATURES 

This model, which relies on characteristics like acousticness, tempo, and speechiness, provides a reasonable level of predictive power. The R-squared value of 0.759 indicates that approximately 75.9% of the variation in song popularity can be explained by these features, showing a strong relationship between a song’s attributes and its success. The Mean Absolute Error (MAE) of 7.98 suggests that, on average, the model’s predictions deviate by about 8 popularity points. This level of accuracy makes it a useful tool for understanding the impact of song features on popularity, though there is still room for refinement.

MODEL 2: PREDICTING POPULARITY BASED ON GENRE, ARTISTS, RELEASED YEAR, AND NUMBER OF SONGS

To further refine predictions, we built a genre estimation, assigning a genre to each song based on its audio features and release year.

Example: Identifying EDM/House Songs

• Released after 1980
• Tempo (BPM) falls within 120-160
• Danceability is higher than 0.65 (65%)

The Genre + Artists + Num Songs model performs significantly better. This could partially explain that factors like who the artist is and what genre the song belongs to have a major influence on popularity. Listeners often gravitate toward familiar artists and genres they already love, making these aspects more impactful than individual song characteristics.

While features like acousticness, tempo, and speechiness contribute to the overall listening experience, they don’t dictate a song’s success as much as an artist’s reputation or the genre’s preference.

From a statistical perspective, using these features to the model increased its R-squared value to 0.805 and lowered the Mean Absolute Error (MAE) to 6.84, making it a more reliable predictor than models based solely on song features. This means the model explains over 80% of the variation in song popularity, demonstrating that an artist’s established presence and genre affiliation play a crucial role in determining success.

Random Forest Regressor (R² = 0.809, MAE = 6.74)

provided a more accurate prediction of song popularity by capturing complex, non-linear relationships between audio features. Unlike traditional linear models, this approach leverages multiple decision trees, each analyzing different subsets of data, and averages their predictions to reduce errors and overfitting. With an R-squared value of 0.809, the model explains 80.9% of the variation in song popularity, making it a strong predictor. The Mean Absolute Error (MAE) of 6.74 indicates improved precision, with lower deviation in predictions compared to the linear regression model. The most influential factors in determining popularity included acousticness, speechiness, loudness, and duration, suggesting that both the texture of a song and its structural elements play a significant role in listener engagement.

Logistic Regression 

The Logistic Regression model achieved approximately 78% accuracy in predicting whether a song would be popular or not. This classification model identified loudness, valence, and danceability as the most influential factors in determining success. Loudness reflects the song’s intensity, valence measures its emotional positivity, and danceability indicates how rhythmically engaging it is. Together, these features help define a song’s overall appeal, making this model useful for distinguishing between hits and less popular tracks.

Key Takeaways for Artists and Record Labels

• Emotion and energy matter: Listeners gravitate toward high-energy, emotionally powerful songs.
• New music has an edge: Fresh tracks have a higher chance of success due to Spotify’s algorithmic preferences.
• Genre plays a role: Pop remains dominant, while older genres struggle to maintain relevance in streaming culture.

Final Thoughts

The music industry is increasingly data-driven. By leveraging audio features, machine learning, and historical trends, artists and record labels can optimize their strategies for making a hit song. Whether you’re an aspiring musician or an executive scouting talent, understanding these data-backed insights can help you make better decisions in the world of streaming.