A Robust Classifier for Unbalanced-Data Learning

In actual work, imbalanced datasets are common, such as in credit card fraud detection, disease prediction, spam identification, and other tasks, where the number of positive and negative samples often varies greatly. This imbalance will bring great challenges to traditional machine learning models, as they usually assume that the data is balanced. To address this issue, this paper proposes a new method using deep transfer hybrid models to solve the problem of unbalanced data learning. The method approach (RN50STN) that incorporates the spatial attention mechanism of spatial transformer networks (STN) and the advantages of transfer learning based on pre-trained ResNet50 is proposed. Combined with a newly designed online OARE algorithm, the RN50STN model improves its learning performance on unbalanced data.