The Credit Risk Problem—A Developing Country Case Study

Crediting represents one of the biggest risks faced by banking sector, and especially commercial banks. In literature, there have been a number studies concerning credit risk management, often involving scoring systems making use machine learning (ML) techniques. However, specificity individual banks’ datasets means that choosing techniques best suited to needs given bank is far from straightforward. This study was motivated need Credins Bank in Tirana for reliable customer tool suitable with bank’s specific dataset. The dataset question presents two substantial difficulties: first, high degree imbalance, second, level bias together low confidence recorded data. These shortcomings are largely due relatively young age private system Albania, which did not exist as such until early 2000s. They encountered more conventional feature literature. present therefore has real contribution make existing corpus research on scoring. first important be addressed imbalance. practice, proportion good customers may many times bad customers, impact unbalanced data classification models an element considered. second relates or incompleteness information emerging developing countries, where economies tend function large amount informality. Our objective this identifying most appropriate ML methods handle Bank’s dataset, various tests we performed purpose yielded abundant numerical results. overall finding strength these results kind can dealt using balanced random forest methods.