Loan-eligibility prediction for Airtime Credit Service Subscribers using Machine Learning: The case Ethio Telecom

Ethio telecom is largest telecommunication companies in Ethiopia. Ethio telecom begun airtime credit service in 2010 to its above 70 million mobile subscribers. The service aims to maintain customer relationships, minimize churn, and generate additional income. However, collecting service money from subscribers is a challenge, resulting in losses of around 2.7 million birrs. To address this, a machine learning-based technology solution was developed to predict eligible customers based on historical data usage. The model uses an Ethio telecom dataset with 114871 rows and 18 columns. Five machine learning classification algorithms were selected for the proposed solution based on the research behavior and literature review: Random Forest, Logistic Regression, Gradient Boosting machines, Naïve Bayes, and Support vector machines. These algorithms are implemented on the prepared dataset and evaluated using model evaluation metrics like accuracy, precision, recall, F1-score are applied. In addition, the confusion matrix table and receiver operating characteristics-area under curve are used to evaluate performance. The accuracy of Random Forest was 90.40%, Logistic Regression was 75%, Gradient boosting classifier was 90.42%, Naïve Bayes was 89.8%, and the Support vector machines was 73.1%. We performed comparative analysis between the models to select the robust model. So that, the Gradient Boosting classifier model provide an outstanding result in predicting eligibility for the airtime credit service.