Music Genre Classification

Overview

This project involves building a classification model to predict the genre of a song based on its features.

Process

Data Preprocessing

• Removed rows with missing instance_id, artist_name, and track_name.
• Replaced missing values in duration_ms and tempo with medians.
• One-hot encoded the key column; label encoded the mode and music_genre columns.
• Standardized numerical features.

Train/Test Split

• Split dataset: 500 songs per genre for the test set, remaining for training.

Dimensionality Reduction and Visualization

• Used t-SNE and UMAP for visualization.
• Applied HDBSCAN for clustering analysis.

Model Training

• Added cluster labels as features.
• Used XGBoost for classification.
• Evaluated with ROC AUC and accuracy metrics.

Results

• Mean ROC AUC: 0.9345 (test set).
• Accuracy: 64.59% (training), 58.82% (test).
• Observed overfitting; suggested cross-validation, regularization, and further feature engineering.

Conclusion

• Feature engineering and data preprocessing were crucial.
• Visualized audio features highlighted the complexity of genre classification.
• Further improvements needed for practical application.

For a detailed description, refer to report.pdf.