Revenue Acceleration from electricity account receivables is one of the energy companies' efforts to maintain cash flow so that they can carry out operational activities and carry out investment activities to develop company assets. Factors that influence electricity bill payment behavior include the location of consumers, the amount of the bill, payment point facilities located around consumers' homes, the use of digital technology as a media of payment, as well as consumer awareness and understanding regarding the time limit for paying electricity bills. Therefore, it is necessary to conduct an analysis so that the company can determine a special strategy for customers who have the potential to be in arrears in electricity bills. To get the characteristic of electricity bill payments, several previous studies have used various classification methods of machine learning such as random forest, nave bayes, SVM, CART, etc. to get the best accuracy. In this research, to increase the accuracy of the model, author using the cluster method with the k-means technique and combining it with the eXtreme Gradient Boosting (XGBOOST) classification method based on data on the characteristics of consumer electricity bill payments. In this study also used hyperparameter adjustment with hillclimbing, random search, and bayesian techniques to increase the accuracy of the model. The model simulation carried out in this thesis gives the result that the combination of the k-means cluster with the XGBoost classification and by adjusting the bayesian technique hyperparameters has a much better model accuracy rate with a value of 89.27% and an Area Under Curve (AUC) value of 0.92 when compared to gradient boosting method with an accuracy rate of only 74.76% and an AUC value of 0.75. Based on the simulation results on ULP Panakkukang customer data, it was found that the subsidy category customer group and customers who often experience power outages have a tendency to be in arrears on electricity bills.