Assessment of Class Imbalance Data Handling with Attention-Based Deep Learning Approach for Robust Financial Distress Prediction in Enterprises

Bankruptcy prediction and credit risk assessment are two of the most crucial problems in finance, remaining an advanced area of research in the financial sector. Financial distress is when an enterprise faces financial problems. Standard analytical methods often depend on limited financial indicators and expert opinions, which miss complex non-linear patterns. With the expansion of Artificial Intelligence (AI) and Deep Learning (DL), financial stress testing has experienced a fundamental change, utilizing progressive methods of computation to improve the precision of risk prediction. This study presents an Attention-based Deep Learning Approach for Robust Financial Distress Prediction in Enterprises (ADLA-RFDPE) method to improve the detection of financial risks in enterprise management. Initially, min-max normalization is used to maintain data consistency, and SMOTE is utilized to address class imbalance. Then, the Elephant Herding Lion Optimizer (EHLO) is used for the dimensionality reduction process. For financial distress classification, an Attention-based Gated Recurrent Unit (Attention-GRU) technique is employed, optimized with the Pied Kingfisher Optimizer (PKO) technique. The experimental results highlighted a superior accuracy of 98.68% over existing approaches on an Australian credit dataset.