Diabetes is a metabolic syndrome, and its annual incidence is rising sharply. Using machine learning methods could help in the early detection of diabetes and early treatment to prevent the condition from worsening. Bayesian network is a type of machine learning algorithm based on the probabilistic graphical model (PGM) and the strength of the model is its ability to combine qualitative visual representations with quantitative reasoning. In the study, a Bayesian network general modelling process framework (including network structure learning, parameter learning, model cross-validation, and inference) is proposed for diagnosing the probability of developing diabetes. For the Pima Indians Diabetes dataset, Bayesian networks built by the framework were used to interpret and visualize the interactions between the influencing factors of diabetes. At the same time, We use the thresholds which are statistical averages for the diabetic population as a benchmark to get the probability values for different combinations, several types of high-risk groups are listed. The study can draw the following conclusions: Glucose is the most direct and important judgment index for measuring diabetes, the higher the blood glucose concentration (>145 mg/dL), the higher the risk of the disease (0.6170) may be. Overweight middle-aged people have a high risk of diabetes (0.6527), and if there is a problem with high blood sugar on this basis, the risk of disease (0.7969) will increase by about 15%. Furthermore, the probability of diabetes can be estimated under any given prior conditions, providing a reference for medical diagnosis.