목적: Predicting recovery in idiopathic sudden sensorineural hearing loss (ISSNHL) is crucial, yet established predictive models are limited. This study aimed to develop AI-based models for predicting ISSNHL recovery, utilizing a large, multi-institutional dataset. 방법:A retrospective analysis encompassed 6499 audiometric tests from 1373 ISSNHL patients, presenting within 14 days of symptom onset across three tertiary hospitals. Tree-based AI models including ExtraTrees, RandomForest, LightGBM, and XGBoost were employed to determine their effectiveness. Each model's performance was individually assessed at each hospital, focusing on the most effective model for the combined dataset. Evaluation metrics included area under the receiver operating characteristic curve (AUROC), accuracy, precision, recall, specificity, and F1 score. 결과:The XGBoost model demonstrated superior performance with AUROCs of 0.87 to 0.88 across the hospitals. Other metrics showed accuracy of 0.842 to 0.855, precision of 0.831 to 0.853, recall of 0.831 to 0.900, specificity of 0.797 to 0.851, and F1 scores from 0.841 to 0.851. The combined model across hospitals indicated an AUROC of 0.85, accuracy of 0.793, precision of 0.813, recall of 0.774, specificity of 0.813, and F1 score of 0.793. 결론:This study successfully develops AI models for ISSNHL prognosis, leveraging a large-scale, multi-institutional dataset for enhanced reliability. XGBoost emerges as a robust tool for prognosis prediction. Clinically, these models can be invaluable in initial consultations for ISSNHL treatment, allowing for patient-tailored discussions and paving the way for individualized, prognosis-model-driven treatment approaches, showcasing the potential of AI in personalized medical care within otolaryngology.