Application of Text Analysis and Ensemble Algorithms in Forecasting Companies Bankruptcy

The article explores methods of textual analysis applied by companies, particularly the analysis that uses compiled dictionaries and is based on machine learning that uses marked-up texts. The authors describe why companies can apply textual analysis. As one of the main directions, the authors consider bankruptcy forecasting, namely the existing methods and practices of its application by companies using machine learning methods, as well as methods of implementing these methods. The paper investigates such a superstructure for machine learning as ensemble algorithms. The authors describe the types of ensemble algorithms and the advantage of using them by comparing single classifier models and ensembles of different types trained on the same data. Additionally, the authors investigate the usefulness of using a non-financial indicator in a bankruptcy prediction model by comparing the accuracy results of two trained sets of models: one using only financial indicators and one including a calculated economic policy uncertainty (EPU) index obtained through textual analysis using dictionaries. The research results show a strong increase in accuracy from adding a non-financial index to the bankruptcy prediction model and the advantage of complex ensemble models. The research novelty lies in the use of advanced ensemble algorithms combining different machine learning models and in the use of a multi-country EPU index to demonstrate the benefits of advanced machine learning methods and the positive effect of using a form of textual analysis in predictive models.