Evaluating Class-Imbalanced Data Handling for Enhanced Financial Distress Prediction Using an Attention-Based Deep Neural Network and Heuristic Optimization Algorithms
Annotatsiya
Financial distress prediction remains important for identifying companies at risk, as it reflects both business stability and economic forecasting accuracy. A major challenge in financial distress data is class imbalance. Deep Learning (DL) techniques have gained global acceptance for financial distress prediction due to their powerful nonlinear modeling capabilities, which capture complex patterns in financial data and enhance prediction performance. This paper proposes an Enhanced Financial Distress Prediction method using Attention-based Deep Neural Network and Heuristic Optimization Algorithms (EFDP-ADNNHOA). The method evaluates the effectiveness of data handling in improving the precision and reliability of financial distress prediction models. The EFDP-ADNNHOA framework begins with data pre-processing, including data cleaning and Z-score normalization to standardize inputs. Feature Selection (FS) is performed using the Flying Fox Optimizer Algorithm (FFOA), whereas classification is carried out with the Bidirectional Gated Recurrent Unit with Attention (BiGRU-A). Finally, the Coati Optimization Algorithm (COA) fine-tunes the parameters of the BiGRU-A model. Tested on the Australian Credit dataset, EFDP-ADNNHOA outperformed existing methods, achieving 97.06% accuracy.