This document discusses using machine learning models to help set sustainable interest rates for small businesses in Africa in order to minimize loan losses. The author created a predictive model combining logistic regression and support vector machines (SVM) that identified borrower characteristics predicting risk. While SVM recovered some losses, probability theory was also needed to achieve around 89% loss recovery. Implementing variable interest rates based on this model could have reduced losses by $240,000 over several years. The model's use going forward could continue recovering losses to reinvest in more loans.