| Abstract [eng] |
If one could assess whether a company will have financial difficulties in the near future, it would be possible to help the company in advance or to prevent owners from selling away assets and spending cash. The problem of company’s bankruptcy forecasting is investigated in this paper. By applying random forest machine learning algorithm along with data imbalance controlling technique it was noticed that the model trained with all-months data is more efficient than the ones trained with specific month data, when forecasting one specific month ahead. Also, after investigating the impact of imbalance in the training data set, it was seen that the models, that were trained with higher imbalance, were better at distinguishing healthy companies, while models trained with smaller imbalance in the training data set were more efficient distinguishing companies that are going bankrupt. Based on random forest results, the most significant variables were determined – arrears to VMI, equity capital, net profit, total amount of employees, age of the company. |
