Abstract: Objective: To compare the value of logistic regression model and random forest model in predicting the risk of aspiration pneumonia(AP) in elderly patients with swallowing dysfunction. Methods: A total of 450elderly patients with swallowing dysfunction in Nanjing First Hospital from January 2021 to October 2022 were selected by convenience sampling method. The factors influencing the occurrence of AP were analyzed by univariate and multivariate logistic regression, and logistic regression model and random forest model were established. The receiver operating characteristic(ROC) curve was used to evaluate the prediction efficiency of the two models for AP occurrence. Results: The incidence of AP was 42.00% in 450 elderly patients with swallowing dysfunction. Multivariate logistic regression analysis showed that consciousness disorder, gastroesophageal reflux, nasal feeding or gastrointestinal nutrition, improper posture, CRP, NLR, LER and lowland drinking water test were the risk factors for AP occurrence, while electromagnetic stimulation therapy, oral cleaning standard and swallowing training were the protective factors for AP occurrence(P＜0.05). According to the random-forest algorithm, the top 8 influential factors for the occurrence of AP were the grade of lowland drinking water test, swallowing training, electromagnetic stimulation therapy, gastroesophageal reflux, consciousness disorder, cough weakness, improper posture, and oral cleaning standard. The logistic regression model predicted the occurrence of AP with an area under ROC curve(AUC) of 0.836 and a prediction agreement rate of 0.432, while the random forest model predicted the occurrence of AP with an AUC of 0.938 and a prediction agreement rate of0.712. Conclusion: The prediction efficiency of random forest algorithm is better than logistic regression.Clinical prediction model should be used to predict the risk of AP occurrence, and corresponding targeted intervention measures should be formulated to prevent AP from occurring.