Year of Graduation
Prediction of Bankruptcy for SME's in Light Industry
The research analyzes the definition of bankruptcy in terms of Russian laws. Furthermore, article consists of evaluation of already existent bankruptcy prediction models. It shows the low level of predictive power for such models, not being accurate enough in terms of present data. It highlights the need in adjustment for modern data. For this purpose, there are realized the methods of machine learning, namely random forest and econometrics with logistic regression. However, even with application of modern methods to update the existent models, the level of their accuracy is still moderate on average. Quality of originated models strongly depends on method of its development. In this research, random forest outdoes logistic regression in accuracy of elaborated models. The final model predicts the bankrupts with high precision. For interpretation of results there was used the Local Interpretable Model-Agnostic Explanation. The key factors for prediction are the structure of assets, organizational form of a company and the return on sales.