Early Detection on Company Bankruptcy: a Comparison of Neural Networks and Logistic Regression

Abstract

Detecting firm insolvency at an early stage is crucial for financial analysis and risk management. This study compares the efficacy of two widely used bankruptcy prediction techniques: Neural Networks (NN) and Logistic Regression (LR). We evaluate each approach based on its accuracy, computing efficiency, and interpretability, aiming to identify a suitable predictive model that aligns with specific objectives, data characteristics, and the need for interpretability in financial decision-making. This research indicates that NN provides superior prediction accuracy but is accompanied by increased computing demands and reduced interpretability. In contrast, LR offers more speed, requires fewer processing resources, and provides explicit understanding of variable correlations; however, it may not perform well with intricate and nonlinear data. This study confirms the significance of choosing a suitable predictive model that balances competing demands of accuracy, efficiency, and interpretability in financial decision-making.