Fault identification, classification and location of distribution network are of great significance to clear the faults quickly and ensure the safe operation of the grid. This paper is based on the big data-driven method, and proposes two algorithms: choosing the least PMU measurements and choosing the best buses from the areas, which are based on evaluation of categorical feature and have differences in their buses selection strategies. Using decision tree classifier and random forest ensemble classifier to verify the performance of the algorithms, Algorithm 1 achieves a maximum accuracy of 91% when using only 7% of the buses, and Algorithm 2 achieves a maximum accuracy of 85% when using data from 16% of the nodes. The experimental results show that the algorithms proposed in this paper can classify and locate feeder faults without necessary measurements of all nodes. In the case of low observability, the propossed methods are achievable for fault identification and location, with the advantages of low number of buses measurements and high accuracy.