الفهرس | Only 14 pages are availabe for public view |
Abstract Over the past five decades bankruptcy prediction studies have been intensively conducted. This backs to the importance of assessing firms’ health for management, investors, creditors, and employees. Bankruptcy prediction studies seek to highlight the predictors (especially the financial ratios) that are important in assessing a firm’s situation along with many trials to develop a highly accurate model in predicting bankruptcy. Many models were developed to predict bankruptcy.Earlier, those relying on traditional statistics, and in recent years, the state-of-the-art machine learning techniques have been applied on firms’ bankruptcy prediction. Themajor goal of this thesis is two-fold; the first is to compare the performance of the well-known linear classifiers and some data mining techniques that are already used for predicting bankruptcy against newly developed ensemble techniques namely LightGBM and CatBoost. These techniques achieved high accuracy in many fields, and it is worthy to investigate their accuracy in predicting bankruptcy. The second is to examine the factors that lead to bankruptcy in Poland and financial distress in Egypt to identify the most relevant factors to the Egyptian business companies. Models are compared based on their accuracy measures including Area Under the ROC Curve (AUC), sensitivity, precision, and recall (specificity) in addition to the learning time, which is used to assess the performance of the algorithms under consideration. In this study, it has been found that extensions of gradient boosting techniquesnamely CatBoost and LightGBM outperformed the other techniques in predicting bankruptcy in Poland with mean AUC scores of 94.2 and 93.6. Most of the features recommended by the models belong to profitability and solvency ratios in addition to the quick ratio, operating ratio, gross profitability ratio, and the activity ratio sales to receivables.Also, all the algorithms used show improvement in their evaluation metrics by handling the imbalance in the dataset using SMOTE balancing technique. For predicting financial distress in Egypt, ensemble methods resulted in higher performance compared to the linear classifiers. The two dominant algorithms were again CatBoost and LightGBM. It has been found that financial distress in Egypt is affected mainly by liquidity and profitability ratios in addition to two activity ratios and one cashflow ratio |