The difficulty of medical insurance risk decision-making in practical applications is that the insured person’s illness is uncertain. For this reason, we try to use machine learning coupled with reservoir sampling to establish a dynamic prediction model to assist medical insurance companies in making intelligent risk decisions. Three aspects were specifically made. First, we build a medical insurance risk decision model, and theoretically we obtain optimal decision rules; then, based on historical data with a fixed sample size, we build a framework for intelligent medical insurance static risk decision-making; finally, to improve static intelligent prediction regarding the lag of risk decision-making guidance, the idea of using machine learning coupled with reservoir algorithm to carry out intelligent dynamic risk decision-making is proposed, dynamic sampling is performed on continuously updated data sets, and a predictive model that is dynamically updated over time is established. Take diabetes as an example of the insured disease, the 2017 Tianchi Precision Medicine Competition-Artificial Intelligence-Assisted Diabetes Risk Prediction Data. In view of the high dimensionality and complex types of data feature variables, the machine learning algorithm selected is random forest. Experiments and comparisons on the training set with the same sample size and the same test set show that the effect of the decision model based on dynamic prediction is better than the static prediction model.