MailGuard Pro: Intelligent Multi-Layer Spam Email Detection and Filtering System

Abstract

The exponential growth of electronic mail communication has been accompanied by a proportional surge in unsolicited and malicious emails, commonly referred to as spam. Spam emails represent a significant threat to organizational cybersecurity, individual privacy, and network resource utilization. This paper presents MailGuard Pro, an intelligent multilayer spam email detection and filtering system that integrates classical machine learning classifiers, deep learning-based natural language processing, and heuristic rule engines into a unified ensemble framework. The proposed system employs a hybrid architecture combining Term Frequency-Inverse Document Frequency (TF-IDF) vectorization, Long Short-Term Memory (LSTM) neural networks, a fine-tuned BERT transformer, and a URL/header metadata analyzer fused through a weighted voting ensemble. MailGuard Pro is evaluated on the SpamAssassin Public Corpus, Enron Email Dataset, and TREC 2007 Public Spam Corpus, achieving an overall accuracy of 97.6%, precision of 96.8%, recall of 96.2%, and F1- Score of 96.5%, outperforming existing baselines including Naive Bayes, SVM, Random Forest, and BERT-only classifiers. The system demonstrates real-time processing capability at 1.4 seconds average inference latency with a false positive rate of 0.31%, making it suitable for enterprise-scale deployment.