Enhancing Credit Risk Assessment in Germany: A GAN-Based Approach with Forward-Looking Variables

Background and Aim: Credit consumption has become a cornerstone of modern economies, making accurate credit evaluation essential to minimizing loan default risks. However, existing credit scoring systems face several challenges, including data imbalance, inefficient sample ratios, and the need for more precise indicator weighting. This study aims to enhance credit scoring for German credit card users by addressing these issues and integrating forward-looking variables to improve prediction accuracy and model stability. Materials and Methods: This research utilizes the UCI Statlog (German Credit Data) dataset, employing a preprocessing pipeline that includes normalization, encoding, and data augmentation via Generative Adversarial Networks (GANs) to address data imbalance. The GAN-based model applies SoftMax classification to predict defaults while utilizing principal component analysis (PCA) combined with macroeconomic and industry-specific variables to enhance the adaptability of the model. Results: Compared with traditional oversampling methods, GAN can generate samples that are closer to the true data distribution, thereby avoiding overfitting and data distortion. The GAN-based model significantly improved predictive accuracy, increasing overall accuracy from 74.25% to 87.92% following data augmentation. The integration of forward-looking variables further enhanced model performance, demonstrating the potential of GANs and dynamic economic factors in credit scoring. Conclusion: This study proposes an advanced credit scoring system that, compared to existing models in the German market, effectively alleviates data imbalance and improves prediction accuracy by enhancing and introducing future variables based on GAN. The findings suggest that GANs can serve as a powerful tool in credit risk assessment, particularly in cases where labeled data is limited. Future research should explore the scalability of this approach across various financial risk prediction tasks.