PREDICTING INSOLVENCY IN BRAZILIAN COMPANIES USING MACHINE LEARNING:

An Analysis of the Random Forest Technique

Abstract

Corporate insolvency prediction serves as a valuable strategic parameter to proactively identify financial and managerial risks. Recent approaches to bankruptcy prediction have predominantly leveraged machine learning algorithms, demonstrating superior predictive accuracy compared to conventional methods. This study highlights the effectiveness of the Random Forest methodology in insolvency assessments. Consequently, an exploratory analysis was conducted to evaluate the feasibility of employing a bankruptcy prediction model within the scope of publicly traded Brazilian companies using machine learning techniques. The sample is made up of companies with delisting status on the B3 stock exchange due to insolvency, between 2005 and 2018, a period in line with the enactment of Brazilian legislation on the subject. Evaluation of the model in Brazilian companies revealed an accuracy of 98% in predicting bankruptcies, highlighting the effectiveness of the Random Forest model as a valuable resource for investors, corporate decision makers and interested parties. This research contributes significantly to the discourse surrounding the adoption of machine learning tools in the field of bankruptcy prediction in the Brazilian business scenario.