Optimized DNETCNN: An effective deep learning technique based feature selection and classification of phishing websites
Abstract
Nowadays, phishing attacks pose a severe risk to human daily life and online social networks. Attackers can make users use phishing URLs by masking unlawful URLs as good ones to collect sensitive data and other advantages. Effective ways of identifying phishing websites are immediately necessary to reduce the risks presented by phishing attempts. However, many powerless and minor influence attributes will the neural network system into the overfitting issue in the training stage. This issue typically makes the issue in phishing website detection effective. To address all the prediction challenges, we introduced an innovative Deep Learning (DL) method for predicting phishing websites. We have five steps to follow in this research. First, it takes input data from the dataset and turns the insignificant and original data into numbers using the OneHot method. Then, in a pre-processing step, it normalizes the data using the min-max normalization method. Then, use the Capsule Network (CapsNet) method to get the attributes from the normalized data. Then, we used the Altruistic Whale Optimization Algorithm (AWOA) to choose the best features for predicting phishing sites. We use the darknet Convolutional Neural Network (DNetCNN) method to tell if a website is real or a phishing site after we choose the features. We employed the Lion Swarm Optimization Algorithm (LSOA) to enhance classification accuracy. We gather data from the four phishing-related datasets for the experiment. The simulation outcomes demonstrate that the identification method proposed in this work is very suitable for identifying phishing URLs. It achieved 99.28\% accuracy, and the training time is comparable to other models.