Hybrid SVM–CNN Fusion Architecture for High-Precision Phishing and Spoofing Detection in Web Traffic

The level of sophistication in phishing and spoofing attacks is ever increasing and this becomes a major online security and privacy threat to users. The conventional machine learning and rule-based systems do not possess the ability to handle advanced schemes of deceit, thus creating an awareness of the necessity of more robust detection models. The study presents a phishing website detector based on a high-performing SVM and CNN ensemble model that is trained on an 87-features structured dataset. The SVM component gives very accurate decision boundaries and the 1D CNN gives rich, abstract feature representations thus the model is able to capture the latent behavioral and structural features. The ensemble can also achieve high classification reliability, obtaining 98% accuracies through soft voting fusion which is better than the single stand-alone model of SVM, CNN and Autoencoder. Extensive screening, such as confusion matrix, ROC-AUC, PR curves as well as feature importance visualization exhibit good generalization between different behaviors of phishing. The hybrid architecture proposed provides an effective, scalable and robust solution that will help to make the current web security systems more resilient to the current cyber threats.