HDNE-NR a Hybrid Deep Neural–Ensemble Framework for Robust Fraud Detection Under Controlled Noise Perturbations
Journal of Soft Computing Paradigm · 29.05.2026 · 8 (2) · 158-183
DOI: 10.36548/jscp.2026.2.005 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Soft Computing Paradigm, online erschienen am 29.05.2026. Verfasst von Sumathi K. und 2 weiteren Autor:innen. Die Arbeit nennt 24 Literaturangaben.
Abstract (Originalsprache)
This paper presents HDNE-NR (Hybrid Deep Neural Ensemble with Noise Robustness) in order to obtain constant fraud detection when subjected to controlled feature and label perturbations. The major purpose was to create a robustness-based predictive framework that can retain discrimination in imbalanced and noisy financial data with great intensity. HDNE-NR provides high-level performance when compared to the deep latent representation learning models; this is due to the combination of heterogeneous ensemble meta-classifiers and adaptive weighted stacking. In clean conditions, the framework had an F1-score of 0.933 and ROC-AUC of 0.992. At high noise of features (σ = 0.20) and corruption of labels (ϵ = 0.10), HDNE-NR maintained higher F1-scores (0.820 and 0.815, respectively) and lower decay curves. It is novel in the synergistic combination of any of the noise-aware learning of representations and adaptive weighting of the ensemble. Generally, HDNE-NR offers an uncertain and scalable framework in fraud analytics in the real world.
Zitieren
APA
K., S., K., P., & S., R. (2026). HDNE-NR a Hybrid Deep Neural–Ensemble Framework for Robust Fraud Detection Under Controlled Noise Perturbations. Journal of Soft Computing Paradigm, 8(2), 158-183. https://doi.org/10.36548/jscp.2026.2.005
Vancouver
K. S, K. P, S. R. HDNE-NR a Hybrid Deep Neural–Ensemble Framework for Robust Fraud Detection Under Controlled Noise Perturbations. Journal of Soft Computing Paradigm. 2026;8(2):158-183. doi:10.36548/jscp.2026.2.005
BibTeX
@article{K2026,
author = {K., Sumathi and K., Priyanga and S., Ramesh},
title = {HDNE-NR a Hybrid Deep Neural–Ensemble Framework for Robust Fraud Detection Under Controlled Noise Perturbations},
journal = {Journal of Soft Computing Paradigm},
year = {2026},
volume = {8},
number = {2},
pages = {158--183},
doi = {10.36548/jscp.2026.2.005}
}
EUVIMED