Phishing Detection Techniques: A Review of Rule-Based, Machine Learning and Deep Learning Approaches

Abstract

Phishing, the use of fake websites and messages to trick people into revealing passwords, bank details or one-time passwords, remains one of the most common ways in which online accounts are compromised. In India, the spread of mobile banking and UPI payments has made SMS and messaging-app phishing a daily risk for ordinary users. This review examines the main techniques developed to detect phishing websites and links. It covers list-based methods such as blacklists, heuristic and rule-based methods, content and visual-similarity methods, machine learning on features extracted from URLs and web pages, and deep learning models that learn directly from the characters of a URL. For each family it discusses how the method works, what it requires and where it fails. The literature shows that blacklists are fast and precise but miss newly created sites, that machine learning on lexical URL features detects most phishing links with little computation, and that deep learning models perform best when large training sets are available. All approaches face the same problems: attackers adapt quickly, public datasets age, and reported accuracies are hard to compare. The review concludes that layered defences, combining lists, lightweight classifiers on the user’s device and user awareness, are most effective, and it identifies the need for datasets and models suited to Indian languages and SMS-based phishing.