Short circuit faults usually occur in the damaged insulation lines or line connections, which will cause serious accidents such as fires and explosions. As the power supply distance increases, accuracy of short-circuit fault detection is insufficient and the process is tedious with the traditional analysis method. In order to solve the problems above, the short-circuit fault detection is classified into the two classification problems while the machine learning method is used. The data of the normal state and short circuit fault state are obtained by the short-circuit simulation experiment. Extract four features from time domain, including the average current and so on. By training support vector machine (SVM) using the different combinations of extraction features above, the model is obtained. The accuracy of classification of the test data set by the model is high. The results show that the short-circuit fault detection method based on machine learning is more accurate and robust than traditional analysis methods.