Detection of Turkish Fraudulent Domain Names to Proactively Prevent Phishing Attacks Using A Character-Level Convolutional Neural Network

Phishing attacks, which aim to steal sensitive information by tricking users into fraudulent websites, continue to pose significant financial and operational threats to organizations and individuals. While various studies have focused on detecting phishing websites using different techniques, only a limited number of studies have addressed the proactive detection of attacks through the analysis of domain names. This paper introduces a novel and proactive approach for real-time detection of phishing websites solely based on their domain names by leveraging the characteristics of Turkish domain names. Detecting the legitimacy of a website based solely on its domain name presents a significant challenge due to the limited availability of features that can expose phishing attacks. To address this problem, the paper proposes the use of character-level embeddings and a convolutional neural network (CNN) to develop a specialized classifier. A new dataset consisting solely of Turkish domain names and their corresponding labels is constructed for training and evaluating the CNN model. The experimental results demonstrate that our proposed model achieves an accuracy of 90%, which is highly promising given the intricacies of the problem at hand. Our study contributes to the advancement of proactive phishing detection methods and underscores the significance of considering linguistic and regional factors in phishing attack analysis.