Phishing Detection Using CNN–RNN with Sparrow Search Optimization Algorithm

Phishing attacks remain a significant challenge to cybersecurity, a hacker’s attempt to deceive users into disclosing vital personal information by pretending to be on an authentic platform. This paper provides a novel phishing web site detection system wherein a convolutional neural network–recurrent neural network (CNN–RNN) is utilized in both detecting phishing web sites and detecting phishing email, respectively, with Sparrow Search Optimization Algorithm (SSOA) being used in optimizing the feature selection as well as the model parameters, respectively. The proposed system gathers phishing data. from the Phish Tank database and then does preprocessing steps such as data cleansing, tokenization, and normalization. The extraction features are used according to their CNN–RNN framework, which is sufficient in terms of capturing the spatial features of phishing web pages and the sequential features of phishing emails. In the optimization step, SSOA is used to tune the model more accurately and efficiently. Experimental results indicate that the CNN–RNN-based model attains 99.27% accuracy, 98.63% precision, 98.89% recall, and 99.12% F1-score, which are higher than other conventional methods. Compared with current schemes, including Region-based Convolutional Neural Network (RCNN), Loopy Belief Propagation (LBP), Random Forest (RF), FastText and CNN, and Support Vector Machine (SVM), the given approach always outperforms those approaches that gie better.