Tuberculosis is the main cause of human disease in the world. In order to find the biomarkers for the diagnosis and treatment of tuberculosis， we analyzed the gene expression profiles of peripheral blood mononuclear cells between individuals with active tuberculosis and latent infection by microarray analysis. First， we used ANOVA to filter out the differentially expressed genes and used the random forest model to predict the status of tuberculosis infection associated with the gene. Then， we compared it with the traditional support vector regression model and showed that the random forest regression model had better predictive performance. Finally， we selected the top 10 of the susceptible candidate genes， which were further proved by literature searching. These findings provide an effective reference for identifying different levels of tuberculosis infection or longterm prevention.