Improving Bankruptcy Forecasting Through the Integration of a Resilient Analysis-Infused Classifier and Harnessing the Power of DenseNet121 Convolutional Neural Network Within Deep Learning Frameworks

In the realm of evaluating financial risks, accurately predicting bankruptcy holds paramount importance for maintaining the stability of economic systems. A longstanding challenge in this field revolves around imbalanced datasets, where instances of bankruptcy are notably fewer than those of non-bankrupt cases. Mitigating this issue involves implementing a robust strategy that combines the Synthetic Minority Over-sampling Technique (SMOTE) with DenseNet121 CNN classification, resulting in heightened model accuracy. Financial data often exhibits an imbalance in class distribution, with occurrences of bankruptcy being significantly outnumbered by instances of solvency. This imbalance poses a substantial hurdle for machine learning algorithms, as they tend to display a bias toward the dominant class, leading to suboptimal accuracy in forecasting the minority class (bankruptcy). The primary objective of this study is to leverage DenseNet121 CNN for bankruptcy detection classification. This goal will be accomplished through the application of SMOTE Analysis, specifically designed to address the challenges posed by unbalanced data. The study aims to achieve an anticipated accuracy of 84 percent, employing the classification report and the confusion matrix as visualization tools in the proposed research. These metrics will contribute to a comprehensive understanding of the results and further enhance the reliability of the bankruptcy prediction model.