Assessment and temporal validation of four machine learning models to predict post-craniotomy nausea and vomiting risk and compare clinical utility.
Key Takeaways
- LightGBM achieved the best overall performance with highest accuracy, recall, F1-score, and AUC
- LASSO selected 14 core predictors from 768 craniotomy patient samples for model development
- Random Forest showed marked performance decline on temporal validation, indicating likely overfitting