🎵 Spotify Song Popularity Prediction with LSTM

A PyTorch-based deep learning project that predicts the popularity of songs on Spotify using their audio features. It utilizes an LSTM model to perform regression on sequential-like data (although each song is treated as a sequence of one).

---

📁 Dataset

• Input file: spotify_songs.csv

• Features used:

  • danceability, energy, loudness, speechiness, acousticness, instrumentalness, liveness, valence, tempo

• Target: track_popularity

---

⚙️ Pipeline Overview

1️⃣ Data Preprocessing

• Drop rows with missing values
• Normalize features using StandardScaler
• Split data into train/test sets
• Convert data into PyTorch tensors

2️⃣ Model: LSTM

A custom LSTM regression model:

Input Size: 9
Hidden Size: 64
Layers: 1
Output: 1 (popularity score)

3️⃣ Training

• Loss function: MSELoss
• Optimizer: Adam (lr=0.001)
• Epochs: 15
• Batch size: 64

4️⃣ Evaluation

• Mean Squared Error (MSE) on the test set
• Line plot of training loss over epochs

---

📈 Sample Output

Epoch 1/15, Loss: 152.3496
...
Epoch 15/15, Loss: 43.7221

Test MSE Loss: 38.9315

---

🖼️ Training Loss Visualization

A plot is generated showing loss vs epochs:

You can use plt.savefig("training_loss.png") to save the figure locally.

---

▶️ How to Run

1. ✅ Install dependencies:

pip install torch pandas numpy scikit-learn matplotlib

2. 📂 Make sure spotify_songs.csv is in the same directory.

3. 🚀 Run the script:

python your_script_name.py

---

📌 Notes

• Designed for educational purposes to demonstrate regression with LSTMs.
• Extendable to deeper networks or sequence data if needed.
• Lightweight and easy to customize.