Development of Ensemble SVM–LSTM Model for Phishing Website Detection

• Abubakar L. IBRAHEEM

  Department of Cyber Security Science, Federal University of Technology, Minna, Nigeria

  Author

• John K. ALHASSAN

  Department of Cyber Security Science, Federal University of Technology, Minna, Nigeria

  Author

• Noel D. MOSES

  Department of Cyber Security Science, Federal University of Technology, Minna, Nigeria

  Author

• Suleiman AHMAD

  Department of Cyber Security Science, Federal University of Technology, Minna, Nigeria

  Author

Phishing is a criminal mechanism employing social engineering techniques to exploit human vulnerabilities and technical loopholes. This is to deceive users into divulging sensitive information, which are in turn used for fraudulent activities. Meanwhile, many machine learning approaches have been proposed in the literature, they often struggle with scalability, adaptability to emerging threats and the trade-off between detection accuracy and computational efficiency. This study aims to enhance phishing website detection through the implementation of an ensemble model that integrate Support Vector Machine (SVM) and Long Short-Term Memory (LSTM) networks. The methodology involves collecting a diverse dataset of phishing and legitimate URLs with a total of 343,525 records. This includes phishing URLs downloaded in CSV file format from PhishTank and Kaggle comprising a total of 167,582 records and legitimate URLs downloaded from the UC Irvine Machine Learning Repository and Kaggle consisting of 175,943 records. Static and Sequential features were extracted. 49 features were extracted and 15 were selected as the most relevant features using Recursive Feature Elimination (RFE) and univariate statistical tests. An ensemble architecture integrating SVM and LSTM networks was then trained using the selected features, employing stratified k-fold cross-validation. The results demonstrate that the proposed approach achieves high detection accuracy of 97.58%, precision 93.54%, recall 96.1% and F1- Score 95.78%, outperforming traditional models and various benchmark classifiers. The findings highlight the effectiveness of combining static and sequential features within an ensemble framework to improve the generalization and robustness of phishing detection systems. Exploring deep learning architectures like CNNs, LSTMs, and GANs in a hybrid framework will likely boost detection capabilities against evolving cyber threats.

• FJET_21_2_2

Copyright (c) 2026 FUDMA Journal of Engineering and Technology

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.