The Modern Approach to Predicting Default Risk in Danish SMEs: A Comparative Study of Machine Learning vs. Traditional Financial Models Using Standardized Financial Data

Abstract

This thesis examines the effectiveness of advanced machine learning (ML) methods in predicting default risks among Danish small and medium-sized enterprises (SMEs) compared to traditional financial ratio-based models. Utilizing standardized financial statement data from the Danish Central Business Register (CVR) between 2018 and 2023 the study specifically compares the predictive accuracy of conventional techniques, including Altman’s Z′-score and logistic regression, against modern ML approaches such as the XGBoost classifier and Random Survival Forest (RSF). The findings demonstrate significant improvements in accuracy with ML models, particularly with XGBoost achieving superior performance in precision-recall and receiver operating characteristics metrics. Moreover, incorporating temporal dynamics through RSF notably enhanced the prediction of default timing, thereby providing practical insights beyond what static models provide. Interpretability, crucial for credit risk management, was evaluated through explainability frameworks such as SHapley Additive exPlanations (SHAP) and permutation importance. These tools clarified the decision-making processes of ML models, ensuring compliance with regulatory transparency requirements. Consequently, the study concludes that ML-based methods substantially outperform traditional models in predicting SME defaults. The greater predictive power, along with strong interpretability, presents substantial practical benefits for financial institutions and regulatory bodies in credit risk assessment.