No more black boxes! Explaining the predictions of a machine learning XGBoost classifier algorithm in business failure

This study opens the black boxes and fills the literature gap by showing how it is possible to fit a very precise Machine Learning model that is highly interpretable, by using a novel ML technique, Extreme Gradient Boosting (XGBoost), and applying new model interpretability improvements. In addition...

Full description

Saved in:
Bibliographic Details
Published in:Research in international business and finance Vol. 61; p. 101649
Main Authors: Carmona, Pedro, Dwekat, Aladdin, Mardawi, Zeena
Format: Journal Article
Language:English
Published: Elsevier B.V 01.10.2022
Subjects:
Business failure
Machine learning
Model interpretability
XGBoost
C 45
Model interpretability
C46
Business failure
G33
XGBoost
G32
Machine learning
ISSN:0275-5319, 1878-3384
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This study opens the black boxes and fills the literature gap by showing how it is possible to fit a very precise Machine Learning model that is highly interpretable, by using a novel ML technique, Extreme Gradient Boosting (XGBoost), and applying new model interpretability improvements. In addition, we identify several significant indicators that could assist in predicting business financial distress. The data were collected from the Eikon database from a sample of 1760 French firms (1585 healthy and 175 failing) in 2018. Identifying the leading indicators of business failure is critical in assisting regulators, and for business managers to act expeditiously before a distressed business reaches crisis point. Our results reveal that higher levels of equity per employee, solvency, the current ratio, net profitability, and a sustainable return on investment are associated with a lower risk of business failure. In contrast, a higher number of employees leads to business failure. [Display omitted] •Bankruptcy prediction techniques have recently switched to intelligent machine learning models.•Machine Learning models are often considered a black box due to their complexity.•Model interpretability is a rapidly expanding ML models.•Using XGBoost and model interpretability could assist in predicting business failure.
ISSN:0275-5319
1878-3384
DOI:10.1016/j.ribaf.2022.101649