Abstract: Data mining algorithm in the era of big data needs to be able to efficiently deal with massive data. Traditional classification algorithms take a long time to train a model and classify the test dataset, and the algorithm is difficult to understand. To deal with the problems, a classification method based on dimension reduction is proposed in this paper. The multidimensional classification problem is transformed into multiple 2D projection surface combination by projection, and a density model of the projection surface is trained for classification. Compared with Support Vector Machines (SVM), Logistic Regression (LR), K-Nearest Neighbor (KNN) and other algorithms, the classification method based on dimension reduction has higher training efficiency and classification efficiency without loss of accuracy. The method is easy to implement, so it can be used for real-time application, such as intrusion detection and traffic scheduling.