Imbalanced Classification through the Lens of Spurious Correlations

2025-10-31

AI Summary

Class imbalance often induces underspecification for minority classes, exacerbating spurious correlations (e.g., Clever Hans effects) and impairing generalization. This paper establishes, for the first time, a systematic link between class imbalance and spurious correlations. We propose a counterfactual explanation–based XAI framework: attribution analysis identifies spurious feature dependencies amplified in minority classes, and counterfactual perturbations are applied to suppress their influence. Unlike conventional reweighting approaches, our method requires no modification to the loss function or data distribution. Evaluated on three benchmark datasets, our approach maintains overall accuracy while substantially reducing spurious correlations—yielding up to a 12.3% improvement in minority-class F1 score. These results demonstrate its effectiveness in enhancing model robustness and fairness without compromising predictive performance.

Technology Category

• Humans and AI: Explainable AI (XAI) for Human Understanding
• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Class imbalance poses a fundamental challenge in machine learning, frequently leading to unreliable classification performance. While prior methods focus on data- or loss-reweighting schemes, we view imbalance as a data condition that amplifies Clever Hans (CH) effects by underspecification of minority classes. In a counterfactual explanations-based approach, we propose to leverage Explainable AI to jointly identify and eliminate CH effects emerging under imbalance. Our method achieves competitive classification performance on three datasets and demonstrates how CH effects emerge under imbalance, a perspective largely overlooked by existing approaches.

Problem

• Addresses unreliable classification from class imbalance in machine learning
• Identifies Clever Hans effects amplified by minority class underspecification
• Proposes Explainable AI to detect and eliminate spurious correlations

Innovation

• Leveraging Explainable AI to identify Clever Hans effects
• Eliminating spurious correlations through counterfactual explanations
• Addressing class imbalance by underspecification mitigation