Loan Default Prediction: Classifying Clients using Risk-Sensitive Learning

Abstract

Loans are an important part of a capitalist economy. Current methods of evaluating potential loans are dated and often require underwriters to use basic credit scores (which may be inadequate due to the bad or no credit history of most micro-loan borrowers) and simple formulas to evaluate credit risk. We use an algorithm of risk-sensitive learning used to minimize risks of huge losses that happen with low probability to classify loan applicants. We evaluate this algorithm using both uncorrelated and correlated loan outcomes to determine it’s effectiveness.