Autoencoder Based Feature Selection for Phishing URL Attack Detection in IoT Using Stacked Autoencoder (AFS-SAE)

With the rapid rise of the economy and technology, phishing assaults have increased in recent years. Phishing can also be done with publicly available sources by attackers with less technical knowledge. Businesses and customers both suffer financial losses, lowering their trust in e-commerce. As a result, countermeasures must be implemented to prevent phishing attempts on the website. Phishing is a type of cybercrime when an attacker targets a specific person or an organisation by elevating or promoting them as an official person of the organisation via an organization e-mail, fake websites, social media, or other communication means. In a phishing assault, the attacker sends phishing e-mails with bogus links or attachments that can perform a variety of tasks, such as capturing the victim’s login credentials or stealing any other crucial data [28]. These e-mails cause victims to lose money through the internet. Phishing is a harmful rising threat on the internet in which attackers attempt to steal the user’s login credentials such as usernames, passwords, or even bank details by exploiting pirated websites. There are some current security mechanisms, such as firewalls, but they are ineffective in blocking many types of attacks since they depend on a set of established criteria that only identify and block predetermined attacks. We need a defence system that can learn attack possibilities so that it can identify and block not just current attacks but also future attacks. Phishing website detection and prevention at the network layer, the proposed work uses a enhanced Dragonfly optimization technique for feature selection to train an Recurrent Neural Network (RNN) model. The dataset, which comprises of a collection of legal and phishing websites, is used to train the model and compared the results using the accuracy, recall, precision, and F1-score criteria.