Hybrid Optimization Enabled Squeeze Net For Phishing Attack Detection

Abstract

Nowadays, phishing attacks are considered as one of the most common cyber attacks which generally harm the security of various internet and communication systems. The phishing websites are produced with the intention of determining the user's financial information by creating cyber threats. The loss of valuable user assets occurs due to the frequent creation and circulation of fake websites all over the internet. The negative impacts of phishing websites include financial loss, loss of intellectual property, reputational harm, and interruption of daily operations. To identify and lessen these attempts, several anti-phishing strategies have been put forth during the past ten years. However, they are still inaccurate and inefficient. To overcome this problem, intelligent approaches, like Deep Learning (DL) is used, which can accurately learn the inherent characteristics of the websites and identify phishing websites. In this paper, the phishing websites are determined effectively using the Fractional Dingo Hunter Prey Optimization-SqueezeNet (FDHPO-SqueezeNet) technique. The different features, like web features, ocular features, and Natural Language Processing (NLP) features are extracted separately from the website data. Later, the optimal features are selected, fused, and augmented, and allowed for the detection of phishing websites using SqueezeNet. Finally, the phishing detection is performed using SqueezeNet, where the detection performance of the SqueezeNet is increased by training using the FDHPO technique. The investigation results revealed that the designed FDHPO-SqueezeNet technique recorded greater performance when compared with other prevailing phishing detection approaches with a maximal of 93.05% accuracy, 94.26% sensitivity, and 93.75% specificity, respectively.