A Boruta-LightGBM Model-based Method For Detecting Electricity Theft

Abstract

Electric theft has great harm to the security and effectiveness of the national grid, so it is very important to detect electric theft behavior of users timely and accurately. In order to improve the accuracy of electric theft identification, with the support of big data analysis, this paper proposes a detection method of electric theft behavior based on Boruta-LighTGBM model. It obtains characteristics through electricity data, adopts SMOTEENN mixed sampling to balance load data, and selects important features by Boruta. It was put into LightGBM integrated learning to train the model, and the test set was used to verify that the model has a high accuracy rate, recall rate, F1 score and AUC value, which provides a fast and effective detection method for theft spot detection.