Existing machine learning based user electricity theft detection methods have insufficient detection efficiency and accuracy. Thus an ISSA-SVC model based on the improved sparrow search algorithm (ISSA) to optimize the parameters of support vector classification (SVC) is proposed. First, the model analyzes the correlation between line loss rate and electricity theft, metered electricity consumption and electricity theft, metered electricity consumption and line loss, electricity supply and electricity theft, electricity consumption on the most recent day and adjacent days, and electricity consumption curves of the users with similar characteristics on each day of the station area to extract the characteristics of the users’ electricity theft. Secondly, it uses the dynamic time warping (DTW) algorithm to calculate their correlation coefficients. Finally, it uses ISSA to optimize the SVC penalty parameter C and kernel parameter g, and detect electricity theft users in the station area. Simulation examples and analysis of real grid data show that the proposed method has higher efficiency and accuracy than traditional power theft detection methods.