We propose a clustering algorithm and a classification algorithm for prisoners. Considering that prisoners feature continuous changing and diversity, we use the hidden Markov model as the clustering model and the latent Dirichlet allocation (LDA) topic model as the classification model, and perform clustering and classification on the management information of various prisoners collected by the emergency command management platform. Experiments show that the hidden Markov model can reflect the temporal changes of each prisoner's performance in the entire sentence, so accurate clustering results can be obtained. The LDA topic model can accurately determine the class of the prisoners according to a variety of their attributes.