- Getting track information from Spotify using the Spotipy library. Link: https://github.com/spotipy-dev/spotipy
- An attempt at Music Emotion Recognition for the kaggle competition @ https://www.kaggle.com/competitions/kcl-artificial-intelligence-competition-22-23
This project uses the Spotipy library to fetch track features from Spotify playlists categorized by mood.
The generated dataset consists of track features that describe the audio properties of songs. Here’s what each column means:
⚡ energy – Intensity of a song (0 to 1). More energy = more power!
🎤 liveness – Is it live or studio? Higher values mean a live audience presence.
🏃 tempo – Beats per minute (BPM). Faster songs = more movement!
🗣️ speechiness – How much speech is in the track?
🎸 acousticness – Higher means more acoustic, lower means more electronic.
🎶 instrumentalness – Measures the presence of vocals. Close to 1? Probably an instrumental.
💃 danceability – Can you groove to it? Higher values mean it’s dance-floor-ready!
⏳ duration_ms – Track length in milliseconds.
🔊 loudness – Measured in dB, higher is louder.
😊 valence – How happy or sad a song feels (0 = sad, 1 = happy).
🎭 mood – The mood label (happy, sad, angry, chill).
Random Forest Classifier (RFC): n_estimators=100, max_depth=None, random_state=42
XGBoost Classifier: n_estimators=100, max_depth=6, learning_rate=0.1, objective='multi:softmax'
Naïve Bayes Classifier: GaussianNB() from sklearn.naive_bayes
XGBoost performs better, achieving a more balanced class distribution and fewer misclassifications.
Random Forest, while effective, struggles with certain classes, leading to a higher rate of incorrect predictions.
If prioritizing accuracy and balanced performance → XGBoost is the better choice.
🔸 Class 0: Represents Angry or high-energy, intense emotions.
🔸 Class 1: Represents Happy or upbeat, positive emotions.
🔸 Class 2: Corresponds to Sad or melancholic emotions.
🔸 Class 3: Represents Chill or relaxed, soothing emotions.
