Design and Development of Machine learning based Enterprise financial crisis Early Warning System

Financial crises have historically led to significant economic downturns, underscoring the need for effective early warning systems. Traditional models, such as logistic regression, typically have difficulty representing complex, nonlinear dependencies inherent in financial indicators, resulting in reduced prediction reliability. To address these limitations, this study proposes an early warning model utilizing a Backpropagation (BP) Neural Network (NN) for enterprise financial distress prediction. The BPNN is adept at modelling intricate nonlinear patterns, making it suitable for forecasting enterprise financial crises. The model is trained on the Jordà-Schularick-Taylor Macrohistory Database. This dataset includes variables such as credit growth, GDP, inflation, and interest rates, providing a comprehensive foundation for crisis prediction. Upon evaluation, the BP Neural Network model attained an accuracy of 98.5%, precision of 97.5%, recall of 97.9%, F1-score of 97.7%, and an AUC-ROC score of 0.989. These outcomes indicate a significant improvement over traditional models, highlighting the model’s capability to serve as a robust early warning system for financial crises.