Veri Mahremiyeti Y ntemlerinin Kredi Risk Siniflandirmasi zerindeki Etkisi

The adoption of centralized machine learning in finance is limited by data privacy concerns and regulations. This study evaluates the impact of federated learning, synthetic data generation, and anonymization on classification performance using the Default of Credit Card Clients (DCCC) dataset. Experiments with four classification models assess the effects of privacy-preserving techniques. Results show that federated learning and synthetic data generation outperform anonymization in accuracy. Notably, models trained on synthetic data achieve performance comparable to or exceeding centrally trained models (highest accuracy: 80.2%, highest F1-score: 65.87%, Support Vector Machine model). These findings highlight federated learning and synthetic data as effective, privacy-preserving solutions for financial applications.