Electricity Bill Recovery Risk Prediction Based on Machine Learning Algorithm

In order to reduce the risk of electricity fee recovery for power companies and improve the accuracy of electricity fee recovery risk prediction. This paper firstly analyzes and collates the existing electricity rate recovery data comprehensively, and obtains a large number of historical electricity rate payment records and user credit information from the power company database. In the process of index selection, the key indicators that are highly related to the risk of electricity cost recovery are selected by combining the expert experience and data characteristics. Based on these key indicators, the paper constructs a machine learning algorithm model, compares the performance of multiple algorithms, and selects the algorithm with the best performance in prediction accuracy and generalization ability. Experiments show that LSTM neural network model is superior to logistic regression model in hit rate, coverage rate, promotion degree, accuracy rate, recall rate and F1 value, showing higher prediction accuracy and effectiveness.