Авторы:
Издательство:
Дата издания
15.05.2026На каком языке издана
ISSN
21921962Область исследования
Аннотация (краткое описание)
Imbalanced learning (IL) poses a major challenge in fraud recognition especially in financial datasets with extreme class imbalances (IR>20:1), where existing sampling techniques such as oversampling and under- sampling fail to address both inter-class and intra-class imbalances. The practical usefulness of current ensemble approaches is limited by their high processing costs or sub-optimal performance. This paper presents the duple-balanced ensemble (DUBE), a novel framework that integrates adaptive resampling with ensemble learning to mitigate inter- and intra-class imbalances simultaneously. Assessed on a real-world financial dataset (n=8,345 transactions, IR=21:1) using Python-based experiments, DUBE was compared with five benchmark classifiers (multilayer perceptron, k-nearest neighbor, bagging, adaptive boosting, decision tree). It achieved a 12% enhancement in F1-score over baseline models with optimal ensemble size of 10, demonstrating robustness to extreme imbalances without relying on computationally expensive distance-based methods. In spite of its robust performance, DUBE may have its efficacy limited by extreme class sparsity or dynamic fraud patterns due to concept drift. These findings highlight DUBE’s potential as a scalable, efficient solution for fraud detection, with future applications in other high-imbalance domains like medical diagnostics and intrusion detection.DOI
10.22967/HCIS.2026.16.028Тип публикаций
Квартиль
Источник квартиля
SCImago (Scopus)Impact Factor
3.9Количество цитирований
0Данный документ располагается в коллекциях
Международные публикации ученых SU
Scopus
Файлы документа
Документ не загружен