Predicting Corporate Bankruptcy Using Machine Learning Models

This research delves into the development and evaluation of machine learning (ML) models tailored for bankruptcy prediction classification tasks. Its objective is to offer insights and recommendations for assessing company efficiency while employing analytical approaches for interpreting business data. To optimize the feature selection process from among 96 attributes in the Taiwanese Bankruptcy Prediction dataset, the model-based feature selection approach using the Random Forest classifier is applied. After determining that the specific set of 15 attributes should be used for training ML models, five models are chosen to address the task’s complexities: the Support Vector Classifier, Logistic Regression, Decision Tree Classifier, K Nearest and Random Forest Classifiers. Among these models, the Random Forest Classifier excels with superior overall efficiency, boasting an accuracy of 0.972 and a processing time of 1.292 s, establishing itself as the optimal choice for corporate bankruptcy prediction when dealing with an unbalanced dataset. This research also successfully demonstrates the applicability of other tested ML models in solving classification problems, thereby indicating the reliability of predictions for each of them.