Förutsägelse av konkurser : En jämförande studie av maskininlärningsmodeller och metoder för att hantera obalanserad data

Predicting bankruptcy is an important task in financial risk management, but is challenged by the inherent class imbalance presented by real-world data. This study evaluates the performance of three machine learning models: Logistic Regression (LR), Support Vector Machines (SVM), and eXtreme Gradient boosting (XGBoost), for bankruptcy prediction. The models were trained on a dataset consisting of financial indicators from US companies. To address the class imbalance, three resampling methods, SMOTE, DBSMOTE, and random under-sampling (RUS), were applied and evaluated for their impact on model performance. The results showed that XGBoost consistently outperformed the other models in most metrics, specifically in accuracy, specificity, balanced accuracy, and Area Under the Curve (AUC) score. Among resampling methods, no single technique was superior across all models. However, RUS yielded particularly strong results with XGBoost.