EDEL: Error-Driven Ensemble Learning for Imbalanced Data Classification

Xu Sun, Qiang Gao, Li Huang, Zixuan Qin, Yuexian Wang, Shun Zhang

Keywords: Imbalanced Data Classification, Machine Learning

TL;DR: EDEL, which dynamically introduces misclassified instances to enhance model performance, particularly in extreme imbalance, as validated by extensive experiments on real-world datasets.

Abstract: The class imbalance problem poses a critical challenge in high-stakes applications such as fraud detection, where the minority class often represents rare but consequential cases. In such settings, misclassifying minority instances can lead to substantial financial loss, underscoring the need for learning algorithms that remain reliable under severe imbalance. While deep learning methods have achieved remarkable success across various domains, their effectiveness often depends on large-scale datasets, and their black-box nature limits their interpretability, which is a critical requirement in high-stakes scenarios. To address this gap, we propose **E**rror-**D**riven **E**nsemble **L**earning (**EDEL**), an adaptive machine learning algorithm that dynamically introduces misclassified instances during training, thereby placing greater emphasis on hard-to-classify samples. Through theoretical analysis and extensive experiments on multiple real-world datasets, EDEL demonstrates strong effectiveness, particularly under challenging imbalanced conditions.

Primary Area: other topics in machine learning (i.e., none of the above)

Submission Number: 8727