Construction of a corporate financial distress prediction model based on F-DNNs

Abstract

The occurrence of corporate financial distress usually requires a certain period of accumulation, and its early prediction can effectively prevent companies from falling into financial crises. However, current mainstream prediction methods overly rely on the experience of leaders, resulting in unreliable prediction outcomes. This study presents a corporate financial distress prediction model built on Fused Deep Neural Networks. By integrating the advantages of Long Short-Term Memory and Longformer, the model efficiently captures both financial data features and semantic information from textual data. Experimental results indicate that the model achieves 96.7% accuracy, 93.5% recall, and an Area Under Curve value of 0.886, significantly outperforming the comparison models. In practical corporate distress prediction, the model occupies a minimum memory of 645MB with a response time of 76 ms. Meanwhile, its prediction accuracy reaches 92.86%, and the longest early warning time can be advanced by 8.8 weeks, significantly surpassing comparative models. The above results indicate that the model proposed in this paper demonstrates excellent accuracy and applicability in corporate financial distress prediction, effectively addressing the lack of accuracy and reliability in current prediction methods. It provides new ideas and approaches for corporate financial distress prediction, contributing to the advancement of intelligent and efficient development in this field.