Authors: Gandhi Himani Dipak, Prof. C. A. Kolse, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Email remains one of the most widely used channels of communication in personal, academic and business environments, and for the same reason it is a favoured route for spam, phishing, sender spoofing, malicious links and malware attachments. Rule based filters and blocklists depend on known signatures and struggle against new and evolving campaigns. This paper reviews Machine Learning (ML) approaches that analyse email content, header information, sender behaviour and embedded links to classify messages as legitimate, spam or phishing. The review traces the development from Bayesian spam filters and support vector machines to feature-rich phishing classifiers, URL-based detectors and transformer language models, and it explains how sender authentication standards such as SPF, DKIM and DMARC complement learned models. Text preprocessing, feature extraction, algorithm selection with Naive Bayes, SVM, Random Forest and Logistic Regression, and evaluation with precision, recall and false-positive rate are discussed. A layered architecture for an ML-based email security gateway is proposed, together with a research gap analysis, objectives and a comparison of algorithms. The review concludes that combining authentication checks, multi-view ML classification and user-facing warnings provides a smarter and more reliable defence for digital communication, provided that models are regularly updated against emerging threats.
DOI: https://doi.org/10.5281/zenodo.24220481
