Phishing Attack Detection and Secure Data Transfer Using Echo State Networks and Federated Identity Management

Phishing attacks involve fraudulent sites where a hacker steals sensitive information. This paper aims at an advanced phishing detection framework by combining the Echo State Network and Long Short-Term Memory networks which are based on advanced deep learning and machine learning models, in order to detect phishing attacks that use OpenPhish with essential features like domain names, URL lengths and suspicious TLDs. Preprocessing of the data includes standardization for enhanced model performance, whereas the feature selection uses Sand Cat Swarm Optimization (SCSO) in order to get the most relevant features. Autoencoders are used in extracting features to compress the raw features into compact representations. In the case of training the hybrid model ESN + LSTM on the phishing data set, several evaluation metrics including Precision, Accuracy, Recall and F1-score are taken for performance measurements. Results are, the proposed model is able to achieve high accuracy in detection as 99.62% and precision as 99.53% and recall as 99.70%, with an F1-score of 99.62%. The result proves that the technique outperforms existing techniques, such as RF, ET and VC, and the proposed model is capable of robust performance especially for minimizing false negatives making it highly efficient in real-world applications for phishing detection.