An Improved ensemble deep learning model based on CNN for malicious website detection

Abstract

A malicious website, also known as a phishing website, remains one of the major concerns in the cybersecurity domain. Among numerous deep learning-based solutions for phishing website detection, a Convolutional Neural Network (CNN) is one of the most popular techniques. However, when used as a stand-alone classifier, CNN still suffers from an accuracy deficiency issue. Therefore, the main objective of this paper is to explore the hybridization of CNN with another deep learning algorithm to address this problem. In this study, CNN was combined with Bidirectional Gated Recurrent Unit (BiGRU) to construct an ensemble model for malicious webpage classification. The performance of the proposed CNN-BiGRU model was evaluated against several deep learning approaches using the same dataset. The results indicated that the proposed CNN-BiGRU is a promising solution for malicious website detection. In addition, ensemble architectures outperformed single models as they joined the advantages and cured the disadvantages of individual deep learning algorithms.