Cena za vysvětlitelnost: srovnání transparentních a neprůhledných modelů úvěrového skóringu v regulatorním kontextu EU

The development of artificial intelligence is transforming financial services and increasing the efficiency of lending. The European Union has introduced the Artificial Intelligence Act (AI Act) to regulate the market and protect citizens' fundamental rights. Within this framework, credit scoring systems are classified as high-risk AI systems. A fundamental challenge in this context is the relationship between predictive accuracy and regulatory requirements. The regulation imposes increased demands on transparency, interpretability of outputs, and human oversight. While fulfilling these requirements for complex, opaque (black-box) algorithms necessitates integrating additional explainability methods and potentially incurring higher costs to ensure regulatory compliance, traditional, transparent (white-box) models meet these demands more naturally. The aim of the thesis is to compare modelling approaches in the context of new regulatory requirements through a comparative experiment on secondary data, and to evaluate the suitability of transparent (white-box) models and opaque (black-box) algorithms in terms of their technical performance, economic consequences, and regulatory defensibility .