Bridging the Credit Gap with Fintech Lending: Utilizing Mobile Data and Machine Learning Algorithms for Unbanked Populations

This research contributes to the problem of financial inclusion in unbanked citizens by proposing a new credit scoring model by combining Fintech with machine/deep learning solutions and mobile data. People without a banking history are not available to 'traditional' credit scoring algorithms, leaving them with no access to critical credit services. We propose an alternative paradigm to predict credit risk using supervised learning and feature engineering applied to mobile phone metadata. We used models like Gradient Boosting Machines, Logistic Regression, Random Forest and the best area under curve (AUC) of 0.93 resulted with Gradient Boosting Machines which means the model will predict better with little computation. This model also had an 88% accuracy in predicting default. Moreover, the feature importance analysis also demonstrated that the information of behaviors had significant effect on the predictions of creditworthiness. With a credit approval rate of 79.5%, our model also indicates striking potential in broadening equitable financial access to underbanked populations. This paper provides a scalable, data-driven approach to credit assessment in fintech, with an eye towards inclusive financial technology.