Song Skip Prediction Using XGBoost: A Machine Learning Approach for Music Recommendation Systems

Abstract: With the rapid evolution of digital music platforms like Spotify and Apple Music, understanding user behavior has become crucial for enhancing music recommendation systems. One key behavioral indicator is song skipping, which reflects user preferences and engagement levels. This study introduces a Song Skip Prediction System using the XGBoost (Extreme Gradient Boosting) algorithm to predict whether a user will skip a song based on playback and contextual features. The dataset, comprising 149,860 interaction records, was preprocessed through encoding, scaling, and feature selection to ensure data consistency and accuracy.

The model achieved an impressive accuracy of 97.27% and an ROC-AUC score of 0.9770, outperforming traditional ensemble methods. Evaluation metrics such as the confusion matrix and ROC curve confirmed its strong discriminative performance. To prevent overfitting and data leakage, techniques like cross-validation and regularization were employed. The trained model was deployed using a Flask backend with a React-based frontend, allowing real-time skip predictions through a user-friendly interface. Overall, this work demonstrates how XGBoost can effectively model user listening behavior, offering a scalable foundation for intelligent and personalized music recommendation systems.

Keywords: Music recommendation, user behavior, song skip prediction, XGBoost, machine learning, user engagement, Flask, React, real-time prediction, personalization