Airtime and Package Credit Risk Prediction using Deep Learning: The Case of Ethiotelecom

Abstract

Mobile credit service is a type of service that allows prepaid mobile subscribers to use telecom services at any time and from any location after running out of balance. When there isn't a nearby top-up option or when there isn't any more money to top up, customers can use the different services that Ethiotelecom provides and help the company (provider) by becoming an additional source of revenue. However, providing credit service comes with its own set of risks for the company because many subscribers do not repay their credit or they are defaulters. In this research we developed a model for airtime and credit risk prediction. To predict airtime and credit risk, deep learning algorithms are used namely Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), TabNet, and Wide and Deep network with two validation techniques 10-fold cross-validation and separate train test. To facilitate this, the dataset is collected from Ethiotelecom. After dataset is collected, relevant features were chosen and preprocessing operations such as data cleaning, integrating, and aggregating were completed. The experimental results demonstrate that MLP (multilayer perceptron) classifier with a 10- fold cross validation technique obtains a higher classification accuracy of 99.00%, Wide and deep network ranked second with the accuracy of 98.22%, TabNet ranked third with accuracy of 97.54%, and CNN ranked forth with accuracy of 97.27%. We also applied machine learning algorithms namely Logistic Regression (LR), Decision Tree (DT) and Support Vector Machine (SVM). Prediction airtime and package credit has numerous advantages for telecom companies, including raising awareness to mitigate risk, preventing financial loss, and increasing revenue assurance