Genetic Algorithm and GAN based Hybrid Model for Bankruptcy Prediction

In the field of bankruptcy prediction, this work focuses on developing a trustworthy prediction model to address issues with data imbalance and the removal of unnecessary information. The majority of bankruptcy datasets have an unbalanced structure and may contain useless data, which could lower the model's effectiveness. Thus, creating trustworthy models becomes necessary and difficult. Here, synthesized data is generated using a Generative Adversarial Network (GAN) in order to balance the datasets. Next, a Genetic Algorithm (GA) is used to pick significant features. This process is compared to other methods such as the Chi-square test and Backward Elimination (BE). Here, predictions are made using four distinct models of bankruptcy: Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), and Logistic Regression(LR). Observations reveal that Random Forest outperforms the other three predictive models in terms of prediction accuracy, while the genetic algorithm, when used as a feature selection tool, produces better results.