Prediction of the Bankruptcy of Slovak Companies Using Neural Networks with SMOTE

Abstract

Although the bankruptcy prediction models can be a stabilizing element on both macro and microeconomic levels, they are rather a domain of academic research than an instrument, widely applied in a business practice. It is especially true if the models are reflecting the conditions of countries of their origin, rather than countries of their intended uses. Besides, few of the models contain inherent flaws, including the absence of a methodical approach addressing this problem of the severely imbalanced representation of bankrupt companies in financial datasets. The article is focused on the use of oversampling with SMOTE (Synthetic Minority Oversampling Technique) algorithm under the condition of extremely imbalanced data sets of Slovak companies. While the model does not provide a single answer in many (if not most) of the situations, it still could be used for the selection of companies for which the more detailed (and expensive) analysis is not required.

Effects of classification, feature selection, and resampling methods on bankruptcy prediction of small and medium‐sized enterprises

“…Moreover, these studies, together with other studies, have shown that the financial situation of companies can be predicted based on financial ratios. These originally applied classification methods were recently replaced by the studies of Antunes et al (2017), Divsalar et al (2012), Geng et al (2015), Kovacova and Kliestik (2017), Mendes et al (2010), Xie et al (2011), andTumpach et al (2020) that applied data mining techniques like decision trees (DTs), LR, support vector machines (SVMs), neural networks (NNs), Bayesian networks, gene expression programming, or random forests (RFs). Xie et al (2011) showed that MDA and SVM classification methods achieve more than 80% accuracy.…”

“…Therefore, the authors recommended further verifying the proposed methodology's effectiveness on a larger data sample. Tumpach et al (2020) developed an NN model and worked with a highly imbalanced data sample that reflected a real-life dataset of companies with an occurrence of bankrupting companies of 0.32%.…”

“…Although the study achieved 99% accuracy and 100% specificity on an imbalanced dataset, the recall and f-score were 0%. This means that no bankrupting companies were detected by the model developed by Tumpach et al (2020). The comparable predictive methodology was used with NN models in the past to predict mergers and acquisitions, where the error rate achieved, when using a split ratio of 70/30, was 34.8% (Anagnostopoulos & Rizeq, 2021).…”

“…On the other hand, only several studies did not improve when any resampling method was applied (Kou et al, 2021). Despite this, some researchers, like Boughaci andAlkhawaldeh (2020) andTumpach et al (2020), concluded that it is challenging to achieve an acceptable prediction performance on large imbalanced datasets as these models achieved relatively poor results in terms of signal-to-noise ratio.…”

Ensemble Model of the Financial Distress Prediction in Visegrad Group Countries

“…Musa [46] proposed a complex prediction model based on discriminatory analysis, logistic regression, and decision trees. On the other hand, Tumpach et al [47] use oversampling with the synthetic minority oversampling technique (SMOTE). In the Czech Republic, Neumaier and Neumaierova [48] generated the prediction model on the principle of MDA.…”