Design and Application of Data Mining Technique for Predicting Airtime Credit Risk: The Case of Nigeria Telecom
ABSTRACT
Airtime credit is a valuable service that allows prepaid mobile subscribers to access telecom services even when their balance runs out, with the option to pay for it later. This convenience has not only benefited users but also served as an additional revenue stream for operators. However, the provision of airtime credit without requiring any guarantee from subscribers has introduced inherent risks, as many users fail to repay their credit, resulting in default. This study delved into the role of data mining in predicting airtime credit risk. Utilizing the open-source data mining tool WEKA, various classification algorithms, including J48 decision tree, Naïve Bayes, Multilayer Perceptron, and Logistic Regression, were employed to identify the most effective predictive model. The experiment used Ethio Telecom’s prepaid subscriber usage data, consisting of 86,024 instances and eleven attributes, for building and testing the algorithms. The evaluation was performed using WEKA’s 10-fold cross-validation and percentage split test options, and the performance of the models was assessed through a confusion matrix, considering accuracy, precision, recall, f-measure, and ROC area. The J48 decision tree model emerged as the best performer, exhibiting an accuracy of 98.5632% and Precision, Recall, and F-measure of 0.986, with a threshold ROC area of 0.996 in the 10-fold cross-validation test. The Logistic Regression model attained an accuracy of 97.1717%. On the other hand, Multilayer Perceptron and Naïve Bayes classifiers achieved accuracies of 96.7622% and 94.6355%, respectively. Notably, the selected classifier generated important rules and parameters that can aid in the airtime credit decision-making process. Data usage emerged as a critical attribute with substantial prediction power. For instance, subscribers with high data usage but low usage in other areas were more likely to become defaulters. Additionally, attributes such as voice usage and topping up channels demonstrated significant predictive power for assessing airtime credit risk.
Design and Application of Data Mining Technique for Predicting Airtime Credit Risk: The Case of Nigeria Telecom, GET MORE COMPUTER SCIENCE PROJECT TOPICS AND MATERIALS