| Abstract [eng] |
This master's thesis examines the reliability of bankruptcy prediction models for companies in the Lithuanian food service sector. In today’s volatile economic environment, the ability to accurately forecast financial distress is particularly important in vulnerable industries such as food services. The central research question addresses whether existing statistical and machine learning models can reliably predict bankruptcy in this sector. The aim of the study is to evaluate the suitability of both traditional and advanced models and to develop a hybrid bankruptcy prediction model tailored to the sector’s specific characteristics. The analysis includes the Altman Z-score, Ohlson O-score, SVM, and GBM models, along with their extended versions (including macroeconomic indicators). Finally, a hybrid model is developed and its performance evaluated using metrics such as accuracy, specificity, precision, F1 score, AUC, and Brier score. The empirical part of the study uses data from 96 Lithuanian companies, half of which had gone bankrupt. It was found that traditional models demonstrated limited predictive accuracy, while advanced approaches—especially GBM and the hybrid model—achieved significantly better results. Notably, the inclusion of additional macroeconomic indicators had a substantial positive impact on prediction accuracy. The study confirms that effective application of prediction models requires adaptation to sector-specific features and consideration of the broader economic context. |
