Deep Learning Approaches for Multi-Class Classification of Phishing Text Messages

Phishing attacks, particularly Smishing (SMS phishing), have become a major cybersecurity threat, with attackers using social engineering tactics to take advantage of human vulnerabilities. Traditional detection models often struggle to keep up with the evolving sophistication of these attacks, especially on devices with constrained computational resources. This research proposes a chain transformer model that integrates GPT-2 for synthetic data generation and BERT for embeddings to detect Smishing within a multiclass dataset, including minority smishing variants. By utilizing compact, open-source transformer models designed to balance accuracy and efficiency, this study explores improved detection of phishing threats on text-based platforms. Experimental results demonstrate an accuracy rate exceeding 97% in detecting phishing attacks across multiple categories. The proposed chained transformer model achieved an F1-score of 0.97, precision of 0.98, and recall of 0.96, indicating strong overall performance.