Design and Development of Financial Distress Warning of Companies using Hybrid Deep Learning Model

Timely prediction of financial distress is a critical task in corporate finance, enabling stakeholders to make informed decisions and mitigate potential risks. Traditional financial distress models often rely on linear statistical techniques or shallow machine learning methods, which may fail to capture the complex, nonlinear interactions between financial indicators. Moreover, many of these models require extensive manual feature engineering and are sensitive to noise in real-world financial data, reducing their generalizability across industries and time periods. To overcome these limitations, this study proposes a Deep learning (DL) - relied approach by utilizing a Convolutional Neural Network (CNN) and Attention maps (AM) for financial distress prediction. CNN and AM are capable of automatically extracting high-level features from multidimensional data, making them suitable for modelling intricate patterns in financial ratios and time-based trends. The approach was trained and examined on the Financial Distress Prediction Dataset, which includes structured data from over 10,000 firms with labelled distress statuses. The proposed CNN approach demonstrated superior performance over traditional approaches, achieving an accuracy of 99.2%, precision of 98.9%, recall of 98.6%, F1-score of 98.7%, and an AUC-ROC of 99. These results confirm the model’s ability to learn complex financial relationships and serve as a robust early warning system for identifying distressed companies. This method not only minimizes manual feature selection but also elevates predictive accuracy, offering valuable insights for auditors, investors, and financial regulators.