A Data-Driven Approach to Bankruptcy Prediction Using Machine Learning

A crucial component of financial risk management, bankruptcy prediction aids in the informed decision-making of investors, companies, and other stakeholders. Accurately forecasting bankruptcy is one of the major obstacles in this field, particularly when the data is unbalanced. This endeavor is further complicated by the fact that bankrupt enterprises (the minority class) are frequently underrepresented in such statistics. The Synthetic Minority Over-sampling strategy (SMOTE) is a strategy used in this work to correct this imbalance and improve bankruptcy prediction utilizing machine learning approaches. We use seven distinct algorithms, such as Random Forest, Gradient Boosting, and MLP Classifier, on a financial dataset in order to evaluate the efficacy of various machine learning models. We make two contributions: first, we use important metrics like accuracy, recall, precision, and F1-score to assess the performance of these models before and after using SMOTE. The findings demonstrate that Random Forest and Gradient Boosting outperform the other models in recall with a score of 0.8 and are especially good at identifying failing companies. Our method's unique feature is the way we combine SMOTE with numerous classifiers to increase the model's capacity for accurate prediction and generalization. The prediction models' performance is much improved by this combination, providing a more trustworthy instrument for financial decision-making.