EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Enhancing URL-based phishing detection using hybrid data balancing and advanced machine learning classifiers

D.T. Chinyio, Olalekan J. Awujoola, Olayinka R. Adelegan, Chat D. Chinyio, Halima Aminu

Journal of Statistical Sciences and Computational Intelligence · 2025

DOI: 10.64497/jssci.109 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Statistical Sciences and Computational Intelligence, 2025. Verfasst von D.T. Chinyio und 4 weiteren Autor:innen. Die Arbeit nennt 26 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Phishing remains a significant cybersecurity threat, demanding accurate and efficient detection systems. This study evaluates six advanced machine learning classifiers Random Forest, J48 Decision Tree, K-Nearest Neighbors (KNN), Naive Bayes, XGBoost, and Artificial Neural Networks (ANN) for URL-based phishing detection, with a focus on overcoming the challenge of imbalanced datasets. To address class imbalance, a hybrid data balancing technique combining SMOTE (Synthetic Minority Over-sampling Technique) and resampling is implemented to improve model generalization and detection performance. The classifiers are assessed on both the original and balanced datasets using key performance metrics: accuracy, Kappa statistic, precision, recall, F1-score, and computational efficiency (training and testing times). Results show that data balancing significantly enhances model performance, particularly in detecting the minority (phishing) class. Random Forest, XGBoost, and J48 emerged as top performers. Random Forest achieved 98.20% accuracy (Kappa = 96.40%) on the balanced dataset, with improved training (1.30s) and testing times (0.03s), and maintained high precision, recall, and F1-scores (0.98) for both classes. XGBoost showed the highest accuracy (98.55%) and Kappa (97.10%), with substantially reduced computational times post-balancing. J48 achieved 96.70% accuracy with minimal processing overhead. The findings highlight that combining robust classifiers with hybrid balancing techniques significantly improves phishing detection accuracy and efficiency. This approach offers a reliable, low-latency solution suitable for real-time deployment, contributing valuable empirical insights into optimal model selection and data preprocessing for intelligent anti-phishing systems.

Zitieren

APA

Chinyio, D., Awujoola, O. J., Adelegan, O. R., Chinyio, C. D., & Aminu, H. (2025). Enhancing URL-based phishing detection using hybrid data balancing and advanced machine learning classifiers. Journal of Statistical Sciences and Computational Intelligence. https://doi.org/10.64497/jssci.109

Vancouver

Chinyio D, Awujoola OJ, Adelegan OR, Chinyio CD, Aminu H. Enhancing URL-based phishing detection using hybrid data balancing and advanced machine learning classifiers. Journal of Statistical Sciences and Computational Intelligence. 2025. doi:10.64497/jssci.109

BibTeX
@article{Chinyio2025,
  author = {Chinyio, D.T. and Awujoola, Olalekan J. and Adelegan, Olayinka R. and Chinyio, Chat D. and Aminu, Halima},
  title = {Enhancing URL-based phishing detection using hybrid data balancing and advanced machine learning classifiers},
  journal = {Journal of Statistical Sciences and Computational Intelligence},
  year = {2025},
  doi = {10.64497/jssci.109}
}
Kostenlos starten →