OBJECTIVE: This study aims to develop a machine learning-based classification model for cognitive impairment (CI) in elderly deaf patients and analyze the contributions of blood indices and hearing characteristics in identifying CI. METHODS: Blood and audiometric data from 833 elderly deaf patients across three NHANES cycles were used to build a classification model with five algorithms: Logistic Regression, Random Forest (RF), XGBoost, Artificial Neural Networks (ANN), and Support Vector Machine (SVM). The optimal model was selected to rank feature importance. RESULTS: The RF model, with an AUC of 0.834, performed best. Key predictors of CI included gender, systolic blood pressure, PTA+3kHz, neutrophil percentage, calcium, 6kHz hearing threshold, glycated hemoglobin, lymphocyte count,etc. CONCLUSION: Hematological markers and hearing thresholds, especially the 3kHz threshold, are significant in identifying CI in ARHL, suggesting the need for further clinical exploration.