By making use of patient similarity, the paper screens study cohort of various sizes, establishes personalized and non-personalized diabetes prediction models based on Logistics Regression(LR), Decision Tree(DT) and Back Propagation(BP) neural network, discusses the difference between the performances of the personalized disease prediction model and the non-personalized one based on patient similarity, as well as the difference between the performances of personalized prediction models based on different machine learning algorithms.