Spotify Genre Classification: Pop vs Classical

This project implements a logistic regression model to classify Spotify songs as either Pop or Classical based on their 'liveness' and 'loudness' features.

Dataset

• The dataset contains 9,386 Pop songs and 9,256 Classical songs.
• Features used: 'liveness' and 'loudness'

Model

A logistic regression classifier was with the following characteristics:

• Learning rate: 0.0001 (found to be optimal after testing various rates)
• Number of epochs: 60

Key Findings

1. Data Distribution:

   • Pop songs are highly concentrated in the high loudness range and spread across liveness.
   • Classical songs are spread across loudness but somwhat more concentrated in the low liveness range.

2. Model Performance:

   • Accuracy: 90.90%
   • Precision: 85.89%
   • Recall: 98.27%
   • F1 Score: 91.66%

3. Decision Boundary:

   • The model relies more heavily on loudness than liveness for classification.
   • There's significant overlap between genres near the decision boundary, especially in the mid-range of liveness.

4. Error Analysis:

   • The model is more likely to misclassify Classical songs as Pop (298 cases) than vice versa (32 cases of Pop songs as Classical).
   • This bias is likely due to the greater spread of Classical songs across the feature space.

Conclusion

The logistic regression model overall performs well in distinguishing between Pop and Classical songs, with particularly high effectiveness in identifying Pop songs correctly.