Power cables play an important role in the power system. Identifying cable faults accurately and quickly is of great significance to ensure the stable operation of the power system. However, in practice, the cable fault data are often sparse compared with the normal data, which greatly reduces the accuracy of cable fault diagnosis. In order to solve this problem, this paper proposes a new cable fault diagnosis method based on particle swarm optimization support vector machine (PSO-SVM) for an unbalanced data set, in which the normal samples are extremely far more than the fault ones. First, for the unbalanced data set, the synthetic minority oversampling technique (SMOTE) is proposed for preprocessing so as to improve the balance of the data set. Second, a smart SVM classification strategy is proposed, in which the parameters are optimized by the particle swarm optimization (PSO) algorithm so as to achieve higher fault diagnosis accuracy. Finally, the effectiveness and accuracy of the proposed PSO-SVM method are verified based on actual power cable data. It is demonstrated that PSO-SVM can identify actual cable faults quickly and accurately.