Advancing Cybersecurity: A Gru-ATT-SVM Approach for Phishing Url Detection

ABSTRACT

The rise of technology has transformed the digital realm into a multifaceted space where people engage in various activities such as banking, shopping, education, and entertainment. However, this shift has also exposed users and businesses to constant threats from cybercriminals, particularly phishers, who jeopardize the safety of online interactions and pose significant risks to global security and the economy. To combat these challenges, many methods for classifying and detecting phishing URLs have been developed. Yet, most existing approaches rely on content-based strategies, which struggle to generalize effectively against new, unseen URLs. To address these limitations, our research focuses on automatically capturing the semantic and sequential patterns inherent in URLs, leading to significant advancements in phishing detection. We introduce a novel approach called GAS, which leverages the power of GRU, Attention mechanisms, and SVM algorithms for robust cyber-attack detection. Utilizing benchmark datasets, our goal is to enhance the detection accuracy of our method across a wider range of phishing URLs. Specifically, we conducted experiments using the University of California Irvine (UCI) repository, which includes 11,055 URL entries, comprising 4,898 legitimate URLs and 6,157 phishing URLs. Our experimental results demonstrate that the GRU-ATT-SVM model outperforms previous methodologies, achieving an accuracy score of 96.23%. This underscores its effectiveness in combating phishing attacks. Through our research, we contribute to strengthening cybersecurity measures and mitigating the pervasive threats posed by cybercriminals in the digital landscape.

Keyword: machine learning algorithms, deep learning algorithms, optimized algorithms, nature-inspired algorithms, hybrid deep learning algorithms.

Index terms: machine learning algorithms, deep learning algorithms, optimized algorithms, nature-inspired algorithms, hybrid deep learning algorithms