Publication detail

The choice of the type of image for graphical processing of input data for corporate bankruptcy prediction using CNN

ŠEBESTOVÁ, M. DOSTÁL, P.

Original Title

The choice of the type of image for graphical processing of input data for corporate bankruptcy prediction using CNN

Type

conference paper

Language

English

Original Abstract

This paper deals with the application of Convolutional Neural Networks (CNN) for the bankruptcy prediction of firms in the Czech Republic. It proposes several variants based on the GoogLeNet architecture that predict the bankruptcy of a company 1 to 3 years in advance. The inputs of the model are financial ratios whose values are converted into several types of images. The various types of images are searched to improve the accuracy of company bankruptcy prediction and the right type of image is found. CNN networks can effectively distinguish between active and bankrupt enterprises. The predictive accuracy of the best proposed model ranges between 85 and 93% (depending on the number of years before bankruptcy).

Keywords

bankruptcy prediction, convolutional neural network, financial indicators

Authors

ŠEBESTOVÁ, M.; DOSTÁL, P.

Released

7. 9. 2021

Publisher

VSB - Technical University of Ostrava

Location

Ostrava

ISBN

978-80-248-4548-7

Book

Financial Management of Firms and Financial Institutions

Pages from

164

Pages to

172

Pages count

9

BibTex

@inproceedings{BUT172817,
  author="Monika {Šebestová} and Petr {Dostál}",
  title="The choice of the type of image for graphical processing of input data for corporate bankruptcy prediction using CNN",
  booktitle="Financial Management of Firms and Financial Institutions",
  year="2021",
  pages="164--172",
  publisher="VSB - Technical University of Ostrava",
  address="Ostrava",
  isbn="978-80-248-4548-7"
}