Purpose/Significance To predict the probability of six common chronic diseases based on features from electronic medical record (EMR) data, including chief complaints, vital signs, ancillary examinations, and history of present illness, so as to provide references for auxiliary diagnosis for primary care physicians. Method/Process Using publicly available EMR data, techniques such as data cleaning and natural language processing (NLP) are employed to create a detailed feature. Models are built by using six machine learning algorithms:logistic regression, naive Bayes, deep neural networks, decision trees, support vector machine, and light gradient boosting machine (LightGBM). The performance of models is evaluated using multiple metrics, including accuracy, recall, and AUC. SHAP values are used to enhance the interpret ability of the models. Result/Conclusion LightGBM shows the best performance and can effectively predict chronic disease risk. Feature analysis reveals that chief complaints, specialty examinations, and history of present illness have a significant marginal effect on model predictions, providing a scientific basis for personalized prevention strategies.