Smart dunning to improve collection ratio in internet service provider using C4.5 algorithm

Abstract

Collection Ratio (CR) is the ratio between the total payments of all customers to the total invoice in the current month. CR is also act as performance indicator of the payment collection division for internet service companies. One method that currently used to improve CR is dunning, namely providing customer with information about the bills with various communication methods including visiting the customer's address. Large number of customers and limited number of collectors are the major obstacle in the dunning process. We propose a classification method to predict potential delinquent customers, so that the expected accuracy of dunning increases, which in turn increases the company's collection ratio. We use the decision tree method with the C4.5 algorithm on the historical data from one internet service provider customers in the Riau Islands province, Indonesia. The classification process produces a decision tree with 5, 885 leaves and 6, 765 tree size. The decision tree then evaluated with 10-folds cross validation that resulting in 78.54 percent accuracy, 0.738 precision, and 0.785 recall. The decision tree has been applied to the dunning process of the company.