Bankruptcy Prediction Models: A Comparative Analysis

This thesis investigates the effectiveness of machine learning models in predicting bankruptcy among Italian companies, leveraging financial ratios derived from balance sheets, income statements, and cash flow statements. By analyzing data from 255,919 companies, including 28,360 bankrupt firms, spanning 1996 to 2023, this study addresses the critical need for accurate predictive tools in financial distress. Traditional models like Logistic Regression, LDA, and QDA are compared against advanced methods such as Random Forest, XGBoost, LightGBM, CatBoost, and Neural Networks. The research highlights the importance of feature engineering and the application of metrics like F1-Score and ROC AUC to evaluate model performance on an imbalanced dataset.