Purpose/Significance To explore the application and predictive accuracy of various models in predicting the risk of atherosclerosis in diabetic patients. Method/Process Based on the biochemical data table from the “Diabetes Complications Warning Dataset” provided by the National Population Health Science Data Center, MATLAB software is used to construct risk prediction models for diabetes-induced atherosclerosis. The models are built by using k-nearest neighbors (KNN), decision trees, backpropagation (BP) neural networks, and Naive Bayes algorithms, and which are subjected to comparative analysis. Result/Conclusion In terms of effectiveness, the predictive accuracy of Naive Bayes algorithm is the highest (61.6%), followed by the decision tree model (58.2%), the KNN model (57.7%), and the BP neural network model (55.9%). The results of the confusion matrix and the receiver operating characteristic (ROC) curve indicate that the Naive Bayes model performs best. When comparing the models in terms of effectiveness, performance and stability, the Naive Bayes model is superior.