Abstract: In the process of software reliability modeling, it is very important to effectively improve the prediction accuracy of reliability, which can guide the reliability test and improve the reliability of military equipment software. Related data is collected from Special Vehicles Software Assessment Center. It applies support vector regression to military equipment software reliability model and compare to other machine learning algorithms model. The result illustrates that the SVR algorithm can improve the accuracy of military software reliability prediction, which is reflected in the following four aspects:root mean square error、mean absolute error、root relative square error、relative absolute error.