EMBEDDING AND WEIGHTING OF WEBSITE FEATURES FOR PHISHING DETECTION

One of the most common cyber threats are phishing attacks. During a phishing attack, attackers use various technical and social engineering tricks to try to lure victims to a phishing website. The website looks like it belongs to a trusted organization but is actually run by the attackers and used to mislead victims into revealing their passwords, credit card numbers, or other confidential information. In this paper, we use discrete descriptive website features to detect whether a website is phishing or legitimate. We create a customized embedding layer specifically designed for these types of features, as well as an embedding weighting mechanism that we later apply. We propose a convolutional neural network-based model for phishing website detection and demonstrate its efficacy on three datasets. With accuracy rates of up to 97.56%, the model performed on par with or better than the current state-of-the-art approaches on each dataset.