Credit Scoring Using Ensemble Machine Learning Methods

The aim of this thesis is to examine the performance of selected decision tree-based ensemble machine learning models and to assess their applicability in credit scoring. In the first part, general principles behind supervised machine learning were introduced. Selected ensemble models were then explained in detail, special focus was placed on decision trees, the building blocks of these models. The standard approach to credit scoring function development was introduced and used to develop a benchmark model. In last section, the ensemble models were trained in several different ways. Their out-of-sample performance was measured and compared to that of the benchmark.