EUVIMEDEUVIMED entdecken
Crossref

EARLY CYBER-ATTACK AND FRAUD DETECTION USING ENSEMBLE MACHINE LEARNING ON NETWORK AND TRANSACTION DATA

Dr..K.Sharmila, Mrs.D.Sabareswari

RCHUB JOURNAL OF COMPUTATIONAL INTELLIGENCE SCIENCE AND ENGINEERING (JCISE) · 01.07.2026 · 01

DOI: 10.65725/jcise/2/3/001 ↗

Kurz gefasst

Fachartikel in der Zeitschrift RCHUB JOURNAL OF COMPUTATIONAL INTELLIGENCE SCIENCE AND ENGINEERING (JCISE), online erschienen am 01.07.2026. Verfasst von Dr..K.Sharmila und 1 weiteren Autor:innen. Die Arbeit nennt 15 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Cyber-attacks and financial fraud result in more than $10 trillion in global losses annually, while traditional rule-based detection systems struggle to respond to zero-day exploits and rapidly evolving threats. To address these limitations, this paper presents an ensemble machine-learning framework for early threat detection based on behavioral features extracted from network traffic and financial transaction logs. The proposed methodology is evaluated using real-world benchmark datasets, including NSL-KDD and UNSW-NB15 for cyber-attack detection (125,973 samples) and a European credit-card transaction dataset for fraud detection(284,807 samples). Data preprocessing involves feature engineering and SMOTE-based resampling to address class imbalance. Detection models are constructed using stacking ensembles of Random Forest, XGBoost, and Gradient Boosting, with Logistic Regression serving as the meta-learner. Experimental results demonstrate that the stacking ensemble achieves strong performance for multi-class intrusion detection, attaining 91.1% accuracy, 94.6% recall, 92.5% F1-score, and an AUC of 0.98, outperforming baseline methods by 12.3%. In the fraud detection task, the Gradient. Boosting model achieves 97.9% accuracy and precision, with an AUC of 0.998, while reducing false positives by 45% compared to traditional rule-based screening systems. Feature importance analysis highlights connection rate spikes, transaction timing irregularities, and protocol diversity as key indicators of malicious behavior. The proposed framework is designed for seamless integration with SIEM platforms, enabling real-time threat scoring and scalable early warning capabilities with minimal analyst intervention. Overall, this work contributes an optimized detection pipeline, practical deployment benchmarks, and actionable behavioral insights that support security operations, advancing the application of machine-learning techniques for defending against evolving cyber and financial threats.

Zitieren

APA

Dr..K.Sharmila, & Mrs.D.Sabareswari (2026). EARLY CYBER-ATTACK AND FRAUD DETECTION USING ENSEMBLE MACHINE LEARNING ON NETWORK AND TRANSACTION DATA. RCHUB JOURNAL OF COMPUTATIONAL INTELLIGENCE SCIENCE AND ENGINEERING (JCISE), 01. https://doi.org/10.65725/jcise/2/3/001

Vancouver

Dr..K.Sharmila, Mrs.D.Sabareswari. EARLY CYBER-ATTACK AND FRAUD DETECTION USING ENSEMBLE MACHINE LEARNING ON NETWORK AND TRANSACTION DATA. RCHUB JOURNAL OF COMPUTATIONAL INTELLIGENCE SCIENCE AND ENGINEERING (JCISE). 2026:01. doi:10.65725/jcise/2/3/001

BibTeX
@article{DrKSharmila2026,
  author = {Dr..K.Sharmila and Mrs.D.Sabareswari},
  title = {EARLY CYBER-ATTACK AND FRAUD DETECTION USING ENSEMBLE MACHINE LEARNING ON NETWORK AND TRANSACTION DATA},
  journal = {RCHUB JOURNAL OF COMPUTATIONAL INTELLIGENCE SCIENCE AND ENGINEERING (JCISE)},
  year = {2026},
  pages = {01},
  doi = {10.65725/jcise/2/3/001}
}
Kostenlos starten →