Abstract—Detecting financial distress in businesses that lead to bankruptcy has been studied for a century. Building large labeled bankruptcy data sets is non-trivial and challenging. We produce an imbalanced data set of bankrupt and non-bankrupt Greek Small and Medium-sized Enterprises (SMEs) covering three years before the bankruptcy data and utilize it to test the bankruptcy predictive ability of well-known statistical and several supervised classifiers. A set of machine learning classifiers has been utilized demonstrating good predictive ability. The AutoML supervised classifier applied to the entire imbalanced data set shows worth noticing performance. We implement several supervised algorithms in a semi-supervised framework to remedy the imbalance of the data set and observed better overall performance than the supervised ones. To measure the effect of combining data from compatible European and Greek markets, we developed customized and AutoML-based transfer deep learning classifiers to predict the bankruptcy of Greek SMEs. Our findings justify transfer learning as an alternative methodology for studying bankruptcy prediction-related problems.
 
Keywords—bankruptcy prediction, statistical models, hazard models, supervised machine learning, self-training, semi-supervised, transfer learning


Copyright © 2024 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.