The Human Factor in Cybersecurity: Can Machine Learning Detect Phishing Attempts More Effectively than Traditional Rule-Based Methods?
Abstract: Phishing emails succeed not because computers are weak but because people can be persuaded. A message that looks like it comes from a bank, a school or a delivery company, and that creates a sense of urgency, can lead even careful users to click a harmful link or reveal a password. The first line of defence is the email filter. Traditionally, filters have relied on fixed rules such as lists of suspicious words and known bad websites. Machine-learning (ML) filters instead learn the patterns of phishing from labelled examples. This paper asks how effectively ML models classify phishing emails compared with rule-based methods. It combines a review of published studies with a small classroom-scale experiment in which a keyword rule filter was compared with three ML models (Naive Bayes, Logistic Regression and Random Forest) on a constructed set of 1,212 emails. On phishing written in a familiar style, all methods performed well, with the rule filter reaching 96.0% accuracy and the ML models 99.6% to 100%. On phishing written in a new, calmer style, the rule filter caught only 8 of 100 phishing emails, while the Random Forest model caught 87 with only one false alarm. The findings agree with the wider literature: ML is clearly more effective and adapts better to new tricks, but it is not perfect, and it does not remove the human factor. A layered defence that combines ML filtering with user awareness is recommended.
Keywords: phishing, social engineering, machine learning, rule-based filtering, email security, Random Forest, human factor
How to Cite?: Manvigas Singh Aulakh, Raghu Raja Mehra, "The Human Factor in Cybersecurity: Can Machine Learning Detect Phishing Attempts More Effectively than Traditional Rule-Based Methods?", Volume 15 Issue 10, October 2026, International Journal of Science and Research (IJSR), Pages: 549-555, https://www.ijsr.net/getabstract.php?paperid=SR261005173921, DOI: https://dx.doi.org/10.21275/SR261005173921