In order to solve the problems that in direct current（ DC） power supply system

accuracy of arc fault detection is insufficient and the process is tedious with the traditional analysis method. The DC arc fault detection is classified into two classification problems while the machine learning method is used. The data of normal state and arc fault state are obtained by DC arc experiment. Extract four features from time domain

including the average current and so on. At the same time

extract three characteristics from frequency domain

such as standard deviation of high frequency component

etc. By training support vector machine（ SVM） using the extracted features above

classification model is obtained. The accuracy of classification of the test data set by the model is94. 483 %

the result proves that this method can be used to detect DC arc fault effectively

improve detection precision

and can be popularized easily because of simple steps.