Abstract: Analyzing and judging high-risk groups is an effective way to improve the rate of solving crimes in public security work. Therefore, this paper attempts to build a method for identifying high-risk groups. Through correlation analysis of previous suspect data and case data, an indicator of whether crimes will continue in the next year was constructed to measure and identify high-risk groups. The index was used as the label, the suspect information and case information were used as input features, and the AdaBoost model was used for training and learning. The obtained model was used to predict the suspects and their cases in the latest year, and the list of high-risk groups was obtained.