Malicious email detection using enhanced SVM and CBAM-efficientnet-BO

ABSTRACT: The increasing sophistication of malicious emails poses a significant challenge to traditional classification methods, as text-only models fail to address the diverse nature of evolving threats, including malicious attachments. This study aims to develop a robust multi-modal email classification framework that integrates textual and visual analysis to enhance detection accuracy. The proposed method employs an enhanced Support Vector Machine (SVM) with TF-IDF and Class Variance (TF-IDF-CV) for textual analysis, while malicious attachments are detected using GGE-based visualization with CBAM-EfficientNet-BO. To improve decision-making, an XGBoost ensemble model combines predictions from both modalities , addressing challenges with heterogenous email structures and obfuscated content. Experimental evaluation demonstrates the effectiveness of the proposed framework, achieving 99.6% accuracy. The results indicate that integrating textual and visual analysis significantly improves the detection of complex email threats, offering a practical solution for modern malicious email detection challenges.