In recent years, machine learning has been widely used in all aspects of social production and life, including transportation, finance, etc., and predicting unknown situations based on existing information is also a common application method of machine learning. Therefore, the topic of prediction problems and machine learning models has always been a common research hotspot and direction of machine learning. However, there are still many problems in the current common prediction models. The prediction method based on the frequency and possibility of data retrieval has insufficient prediction accuracy and high error rate. The specific retrieval method obtained by training in the common model is relatively fixed, which can not change in real time with the update and change of the original data, and the dynamic is poor. At the same time, there is no certain security protection in the prediction, which is easy to cause certain privacy leakage in the interaction process, that is, poor security. In order to solve the above problems, we propose a rule learning prediction model based on private information retrieval (PIR-RL). The model use rule learning to realize the prediction function, and the rule learning can help to extract rule features to achieve the accuracy of prediction. At the same time, inspired by SealPIR, this paper proposes a Target Prediction Private Information Retrieval (TP-PIR) to achieve privacy protection. Among them, the dynamic nature of rule learning and the low computational cost have certain advantages in the face of practical problems. The lightweight TP-PIR also achieves privacy protection while ensuring less computational communication overhead. From the theoretical analysis and experimental results, this model has good stability and practical value, and can realize the coordination of prediction service in privacy protection.