Application of an Ensemble Learning Method in the Risk Prediction of Electricity Tariff Recovery

Abstract

Abstract Electricity tariff recovery has always been an important part of power companies. However, there are often some users who are in arrears in electricity charges. Therefore, how to make effective risk assessment for electricity customers is the key work at present. In this paper, an ensemble learning method named gradient boosting decision tree-random forest-adaboost-logical regression (GRAL) is proposed to construct a risk prediction model of electricity tariff recovery. Firstly, the data provided by the power company was preprocessed with feature engineering. Secondly, the GRAL ensemble learning method and the traditional logistic regression algorithm are used separately to construct the risk prediction model of electricity tariff recovery to predict the probability that the user will default next month. Finally, we make a risk assessment for users according the probability and predict whether the user in different risk levels will be in arrears. Experimental analysis shows that the proposed GRAL ensemble learning method has better experimental results compared with the traditional logistic regression algorithm, this method can effectively evaluate the payment behavior of power customers and make a more accurate prediction about whether users will be in arrears.