An explainable hybrid deep learning-optimization framework for robust phishing attack detection using GAN and transformer-based feature learning

This study proposes to improve accuracy of phishing detection by proposing a new hybrid deep learning framework that combines data augmentation, feature transformation, and optimization-based feature selection. The proposed approach integrates a Generative Adversarial Network (GAN) to generate synthetic phishing samples, followed by feature extraction using a combination of feature extraction using a combination of Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), Fully Modified Residual Convolutional Neural Network (FMRCNN), and Transformer models. To reduce feature dimensionality, the Black-Winged Kite Algorithm (BKA) is applied, while classification is performed using a Support Vector Machine (SVM). Experimental findings on Phishtank dataset demonstrate that the suggested model achieves an accuracy of 98.67%, outperforming other approaches in terms of precision, recall, and F1-score. The novelty of this work lies in the unique combination of GAN with CNN–GRU–FMRCNN architectures for phishing detection, further enhanced by hybrid optimization techniques and interpretability via SHAP (SHapley Additive exPlanations) analysis.