Hit Song Prediction Dataset: Audio Features and Billboard Chart Matches for 515,576 Songs

Eva Zangerle and colleagues from Universität Innsbruck created this dataset for the 2019 ISMIR conference. It combines audio features extracted from the Million Song Dataset with Billboard Hot 100 chart performance data. The dataset covers 515,576 songs representative of western commercial music released between 1922 and 2011.

Use Cases

• Predicting Billboard chart success based on low- and high-level audio features extracted by Essentia.
• Analyzing the relationship between musical characteristics and commercial performance for western commercial music.
• Training binary classification models using Billboard hits as positive samples and non-hits as negative samples.
• Studying temporal trends in popular music features using the provided release year information.

Strengths

• Integrates two distinct data sources: the large-scale Million Song Dataset and authoritative Billboard Hot 100 chart data.
• Provides release year information for 515,576 songs, enabling temporal analysis.
• Includes both low-level and high-level audio features extracted using the standardized Essentia toolkit.

Limitations

• Row count for the full feature set is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect temporal bias inherent to the source, covering music only up to 2011.

Provenance

Source

Derived from the Million Song Dataset and Billboard Hot 100 charts.

Collection Method

Audio features were extracted using the Essentia toolkit; chart matches were identified and compiled.

Time Range

1922 to 2011