EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Few Shot Learning For Ransomware Detection Using Siamese Neural Network

Usman Ahmad Bello, Abubakar Bello Tambuwal, Anas Tukur Balarabe

FUDMA JOURNAL OF SCIENCES · 24.07.2026 · 10 (12) · 85-93

DOI: 10.33003/fjs-2026-1012-5348 ↗

Kurz gefasst

Fachartikel in der Zeitschrift FUDMA JOURNAL OF SCIENCES, online erschienen am 24.07.2026. Verfasst von Usman Ahmad Bello und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Ransomware poses a threat to the availability, confidentiality and integrity of an individual's, organization, or a company’s sensitive data. The attack involves locking digital data and holding its hostage until the attacker receives ransom. Due to the rapid spread of devices and interconnectedness, the increasing prevalence of ransomware attacks is concerning. Several research have been proposed, tested, evaluated and published to improve the methodology and implementation of ransomware detection methods, from behavioral, dynamic to static-based detection. Much research has been focusing on machine learning and deep learning to enhance detection accuracy. However, deep learning, with its advantage over traditional techniques, is hindered by its dependence on large number of datasets, often referred to as "data hunger". An emerging approach called few-shot learning (FSL) method is being used in various domains especially in cyber security due to its effectiveness in cases with limited data. This study investigates the usefulness of FSL in ransomware detection and explores the performance of this model with less data, utilizing its strengths to create a robust, unified model to accurately detect and classify ransomware and good wares. Utilizing the ransomware access pattern (Ransap) dataset converted into grayscale images, the proposed techniques improved the classification performance. Specifically, accuracy increased from 97.69% to 98.33%, precision from 95.71% to 97.52%, and the F1-score from 97.31% to 98.33%. In addition, the model achieved a recall of 99.16% and an AUC of 99.03%, compared with the benchmark values of 100% for both recall and AUC.

Zitieren

APA

Bello, U. A., Tambuwal, A. B., & Balarabe, A. T. (2026). Few Shot Learning For Ransomware Detection Using Siamese Neural Network. FUDMA JOURNAL OF SCIENCES, 10(12), 85-93. https://doi.org/10.33003/fjs-2026-1012-5348

Vancouver

Bello UA, Tambuwal AB, Balarabe AT. Few Shot Learning For Ransomware Detection Using Siamese Neural Network. FUDMA JOURNAL OF SCIENCES. 2026;10(12):85-93. doi:10.33003/fjs-2026-1012-5348

BibTeX
@article{Bello2026,
  author = {Bello, Usman Ahmad and Tambuwal, Abubakar Bello and Balarabe, Anas Tukur},
  title = {Few Shot Learning For Ransomware Detection Using Siamese Neural Network},
  journal = {FUDMA JOURNAL OF SCIENCES},
  year = {2026},
  volume = {10},
  number = {12},
  pages = {85--93},
  doi = {10.33003/fjs-2026-1012-5348}
}
Kostenlos starten →