IoT Phishing Detection Using Hybrid NLP and Machine Learning Models Enhanced with Contextual Embedding

Phishing attacks are currently the most spread type of cyber attacks. A lot of work has been carried out dealing with traditional phishing (Smishing, Vishing, Email phishing, etc.) having its medium, vector, and technical approach approximately identified. The rapid expansion of IoT devices has the users to be surrounded by more connected systems and a higher risk of being phished. In this concern, an effective approach is needed to define these novel forms of phishing attacks and how they can be detected using machine learning techniques. In this paper, we are presenting several experiments in order to cover different IoT phishing attacks. We combined DistilBERT-based log embedding, anomaly detection with Isolation Forest, and a custom ANN classifier. Our study demonstrates a robust pipeline for anomaly detection and classification in network log data.