Interpretable Machine Learning to Predict Diagnostic Yield of Emergency EEG in Altered Consciousness and Seizure Evaluation: The EMINENCE Study

This retrospective study applied machine learning models (Random Forest and XGBoost) to 1,018 emergency department patients to predict whether EEG would show abnormal or epileptiform activity and confirm/rule out initial diagnostic suspicions. Models achieved good performance (AUC 0.79–0.84) and used interpretable methods (SHAP) to identify key clinical predictors for guiding appropriate EEG use in acute settings.