Abstract: Aiming at the problem that the accuracy of the support vector machines （SVM） in detecting abnormal users of electricity was affected by the imbala
nce of the samples and the selection of kernel function, we proposed a diagnosis model of theft and leakage of electricity based on Smote-SVM. Firstly, based on the distribution of users of electricity, the Smote method was used to expand a few samples, and the TomekLink was used to eliminate noise. Secondly, the SVM kernel function was optimized after dimensionality reduction. Finally, the algorithm was applied to the diagnosis experiment of the theft and leakage of electricity for unbalanced users, and compared with the traditional SVM and Smote-SVM algorithms. The experimental results show that the algorithm can significantly improve the detection accuracy of the users of theft and leakage of electricity.

Key words: support vector machine (SVM), unbalance data, diagnosis of theft and leakage electricity