In the development process of the Internet, computer technology and network communication have been rapidly applied. Network security has become a research focus. Based on the discrete cloud intrusion detection method, through sensing a large number of data signals in the cloud environment, some of the interference information is screened, filtered and classified. In this paper, the discrete optimization algorithm and machine learning can better reflect the intrusion information in time. This paper mainly uses the methods of experiment and comparison to experiment the three indicators of SVM and its improved algorithm in intrusion detection. The experimental data show that the accuracy of the improved LE-SVM algorithm can reach more than 95%, and its time consumption is relatively small.