FINNET: A Hybrid Deep Learning Network Analysis and Ensemble Learning Model for Financial Distress Prediction

Abstract

Financial distress prediction plays a crucial role in financial risk management and early warning systems. Traditional models often fail to capture the nonlinear dependencies and intercompany relationships that influence financial health. This study presents FIN-NET, a hybrid system integrating Artificial Neural Networks (ANN), ensemble learning (Voting and Stacking Classifiers), and network analysis for more accurate financial distress prediction. The system leverages K-best feature selection and K-means clustering to extract relevant financial indicators. FIN-NET classifies companies as either 'Financially Healthy' or 'Distressed' and provides explainable insights for decision-makers. The implementation uses Python (Flask, Scikit-learn, NumPy) and MySQL, which ensure modularity, scalability, and real-time prediction. Testing confirmed the system’s robustness, achieving high accuracy across multiple scenarios, making it suitable for academic and financial applications.