Enhancing Oversampling Techniques for Financial Data Imbalance: A Machine Learning Approach to Strengthen Asset Management

초록

A major challenge in the financial domain is the class imbalance problem, which can negatively impact model performance. Although oversampling methods such as SMOTE and ADASYN are commonly used to address this issue, they are often sensitive to outliers, which limits their effectiveness. To overcome this limitation, we propose an improved oversampling approach that integrates advanced outlier detection methods-including Isolation Forest variants and autoencoders-into the ADASYN framework. Experimental results show that our method achieves higher prediction accuracy and more stable performance, making it well-suited for real-world financial applications involving imbalanced datasets.

키워드