A compound-aware encoder-only transformer model for smishing detection

Key Points

• The model achieves a significant increase in detection accuracy for smishing attacks, with a notable 15% improvement over traditional methods.
• Utilizing a novel transformer architecture, it effectively leverages compound features in the text for enhanced detection.
• An observational analysis was conducted using a diverse dataset of actual smishing messages to train the model.
• The approach highlights potential for real-time implementations in mobile security applications against fraudulent messages.