A Deep Learning for Arabic SMS Phishing Based on URLs Detection

The increasing use of SMS phishing messages in Arab communities has created a major security threat, as attackers exploit these SMS services to steal users' sensitive and financial data. This threat highlights the necessity of designing models to detect SMS messages and distinguish between phishing and non-phishing messages. Given the lack of sufficient previous studies addressing Arabic SMS phishing detection, this paper proposes a model that leverages deep learning models to detect Arabic SMS messages based on the URLs they contain. The focus is on the URL aspect because it is one of the common indicators in phishing attempts. The proposed model was applied to two datasets that were in English, and one dataset was in Arabic. Two datasets were translated from English to Arabic. Three datasets included a number of Arabic SMS messages, mostly containing URLs. Three deep learning models-CNN, BiGRU, and GRU- were implemented and compared. Each model was evaluated using metrics such as precision, recall, accuracy, and F1 score. The results showed that the GRU model achieved the highest accuracy of 95.3% compared to other models, indicating its ability to capture sequential patterns in URLs extracted from Arabic SMS messages effectively. This paper contributes to designing a phishing detection model designed for Arab communities to enhance information security within Arab communities.