Music Genre Classification

A machine learning project that predicts music genres from Spotify data using various audio features and metadata.

Overview

This project was developed for the final exam of the Machine Learning course in the Bachelor in Artificial Intelligence. The model is capable of predicting the genre of a song based on features from the music domain (tempo, key, time signature), lyrics domain (speechiness, explicitness), and public perception metrics (popularity, danceability). All data is extracted from Spotify's database.

Features

The model uses the following features for prediction:

• artists: artist names who performed the track
• popularity: track popularity (0-100)
• explicit: whether the track has explicit lyrics
• danceability: suitability for dancing based on musical elements
• energy: perceptual measure of intensity and activity
• key: musical key of the track
• loudness: overall loudness in decibels (dB)
• mode: modality (major or minor)
• speechiness: presence of spoken words
• acousticness: confidence measure of acoustic content
• instrumentalness: prediction of vocal absence
• liveness: presence of audience in recording
• valence: musical positiveness
• tempo: estimated tempo in BPM
• time_signature: estimated time signature (3/4 to 7/4)

Installation

Clone the repository:

git clone https://github.com/sebapsan/music-genre-classification.git
cd music-genre-classification

Install dependencies:

pip install -r requirements.txt

Dataset

Download the Spotify Tracks Dataset from Kaggle: https://www.kaggle.com/datasets/maharshipandya/-spotify-tracks-dataset

Place the dataset.csv file in the data/ directory.

Usage

Training the model

Run the main training script:

cd src
python train.py

This will:

• load and preprocess the data
• build a balanced dataset with 10 random genres (1000 samples each)
• add synthetic missing values to demonstrate imputation
• perform model selection using cross-validation
• fine-tune the best model
• evaluate on the test set

Using the Jupyter notebook

For interactive exploration and visualization:

jupyter notebook notebooks/genre_ml.ipynb

Custom training

it is possible to modify the training parameters in train.py:

df = create_sampled_dataset(
    df_original,
    n_genres=10,              # number of genres to include
    samples_per_genre=1000,   # samples per genre
    random_state=42
)

Model performance

The best performing model achieved:

• Test Accuracy: ~92.65%
• Training Accuracy: ~99.95%

The model uses:

• Classifier: One-vs-Rest LinearSVC
• Preprocessing: StandardScaler, MinMaxScaler, One-Hot Encoding
• No dimensionality reduction
• No sampling (dataset is balanced by construction)

Requirements

• Python 3.8+
• pandas
• numpy
• scikit-learn
• seaborn
• matplotlib
• scipy
• mlxtend
• imbalanced-learn
• missingno
• jupyter

See requirements.txt for specific versions.

License

MIT License. See LICENSE file for details.

Acknowledgments

Dataset: Spotify Tracks Dataset by Maharshi Pandya