If the load in the circuit structure of the household power supply system can be classified and the series arc fault can be detected, the probability of arc fire accidents will be reduced. This paper proposed a K-Nearest Neighbor arc fault detection method based on load classification. In this paper, series arc fault experiments under the condition of six single-load and double-load parallel connection were carried out according to the situation of household power supply circuit. The current of each single load circuit was taken as the analysis object, and the current signal characteristics of the normal area and the fault area were analyzed. The loads were divided into resistance type, motor type and switching power supply type loads. On this basis, an arc fault detection model based on K-Nearest Neighbor was established. The model accuracy is 93.3%. Finally, through comparative analysis, it was verified that the accuracy and generalization of this method in arc fault detection and load classification are good.