Intelligent Corporate Bankruptcy Prediction: A Data-Driven Machine Learning Approach

Abstract

Over the past few decades, corporate bankruptcy has become a source of concern for various business stakeholders such as investors and management. It has also piqued the interest of researchers worldwide. Because there are so many variables that contribute to bankruptcy, it is insufficient to rely solely on a single predictive model; instead, the difficulty is in pinpointing the crucial elements that carry the greatest weight. The significant class imbalance in the data is another significant obstacle that impairs the model's functionality. While many methods, such as Decision Trees, Support Vector Machines, Neural Networks, etc., have been studied before with various pre-processing approaches, we advance this research by applying a novel combination of a Random Forest feature selection technique and BORDERLINE-SMOTE hybrid resampling technique on Polish Bankruptcy dataset. We next apply five classifiers to the altered data: Random Forest, Decision Tree, XGBoost, CatBoost and AdaBoost, and assess the effectiveness of each model. As a result of our suggested approach, provides Random Forest classifier with the highest accuracy of 94% using     optimal prediction based on voting principle.