Abstract: This study proposes an innovative spam email filtering method that enhances the detection of phishing emails while safeguarding privacy-sensitive data in legitimate messages. Unlike traditional approaches relying on Naïve Bayes and machine learning, which prioritize spam classification without addressing sensitive data protection, our method integrates information theory and deep learning towards improving spam email filters, equipping them with privacy-preserving capabilities. We introduce novel higher-order information-theoretic and deep feature vectors, leveraging triplet-based word features to improve model performance. Experiments utilize benchmarking datasets enriched with phishing emails and privacy-sensitive information, enabling robust comparison of classification methods, including Naïve Bayes, Random Forest, Boosting Trees, SVM, Decision Trees, and CNN-based deep learning models. Advanced feature extraction techniques grounded in information theory are evaluated, demonstrating that higher-order features significantly enhance spam and phishing content detection accuracy while reducing misclassifications of privacy-sensitive content.

Keywords: Privacy Protection, Privacy-enhancing Technology, Anti-spam Filters, Phishing emails defense, Information Theoretic Feature Extraction, Message Classification Equivocation, Naive Bayes, CNN, Random Forest, SVM, LSTM, Decision Trees, Bayesian E-mail Filtering, Machine and Deep learning E-mail filtering, Higher order features extraction

Pages: 194-217