Credit risk analysis via machine learning methods: client segmentation based on probability of default

Abstract

The rapid evolution of the technology, the competitive environment, as well as the huge amount of data that is available today, lead businesses to switch to the new digital reality. Automation of processes and decision-making through the use of data using new methods such as artificial intelligence and machine learning are a primary objective of organizations. This interest is strongly present in the banking sector, too. The analysis of the large volume of data that is available, whereas taking into account their personal nature is a huge challenge for financial institutions. Credit risk analysis and assessment is one of the most important processes for this kind of business. In this dissertation, 3 models of supervised machine learning were developed, which classify bank's customers into "good" or "bad" based on the probability of default on their obligations. The algorithms used are Random Forest, KNN and Decision Trees.