Machine Learning Approach to Credit Scoring for Fintech Start-Ups Using Micro Finance Banks in Nigeria

Abstract

In the Nigeria FINTECH markets, lack of recorded credit history is a significant impediment to assessing individual borrowers' creditworthiness, deciding fair interest rates, and exposing this company to humongous risk. Thus, this research compares various machine learning algorithms (logistic regression model alongside Decision tree, Naive Bayes, Adaptive Boosting (Adaboost), Random Forests, K-Nearest Neighbour, and Gradient Boosting methods) on real micro-lending data of LAPO microfinance bank domiciled in Nigeria to test their efficacy at classifying borrowers into various credit categories. The results were validated with test metrics such as confusion matrix, accuracy, recall, precision, and area under the curve, which revealed that machine learning algorithms, can be used successfully to classify new customers into various risk classes, while some machine learning algorithm outperforms the others. Also, risk can be mitigated by ascertaining the creditworthiness of an individual applying for a loan. Finally, customers with no credit history should not be classified as high-risk customers. Generally, Random forest, Decision tree Classifier, and KN Neigbhour machine learning algorithms showed better performance with real-life data than others. The study also demonstrated that classifiers such as random forest algorithms can perform this task very well, using readily available data about customers (such as age, occupation, and gender). This presents an inexpensive and reliable means for FINTECH institutions around the developing world, especially Nigeria, to assess creditworthiness without credit history or central credit databases.