Investigating Group Distributionally Robust Optimization for Deep Imbalanced Learning: A Case Study of Binary Tabular Data Classification
This work addresses the class imbalance problem for deep learning practitioners, particularly in tabular data, but it is incremental as it applies an existing gDRO method to a new data type.
The study tackled the class imbalance problem in deep neural networks by investigating group distributionally robust optimization (gDRO) for binary tabular data classification, finding that gDRO outperformed empirical risk minimization and classical methods in terms of g-mean and roc-auc across various benchmark datasets.
One of the most studied machine learning challenges that recent studies have shown the susceptibility of deep neural networks to is the class imbalance problem. While concerted research efforts in this direction have been notable in recent years, findings have shown that the canonical learning objective, empirical risk minimization (ERM), is unable to achieve optimal imbalance learning in deep neural networks given its bias to the majority class. An alternative learning objective, group distributionally robust optimization (gDRO), is investigated in this study for imbalance learning, focusing on tabular imbalanced data as against image data that has dominated deep imbalance learning research. Contrary to minimizing average per instance loss as in ERM, gDRO seeks to minimize the worst group loss over the training data. Experimental findings in comparison with ERM and classical imbalance methods using four popularly used evaluation metrics in imbalance learning across several benchmark imbalance binary tabular data of varying imbalance ratios reveal impressive performance of gDRO, outperforming other compared methods in terms of g-mean and roc-auc.