Explainable AI driven Phishing Detection in Email, SMS and URL with a Unified System

Phishing posed a continuing risk where bad guys use false emails, messages or links to trick people. The authors unveiled a new tool named PhishGuard and claimed it is a hybrid multi-channel phishing detection system that combines machine learning, rule-based analysis, and explainable AI to improve both the phishing detection accuracy and user trust. The system uses handcrafted thirty lexical, domain, and content-based attributes for URL analysis, which are fed to an ensemble of Random Forest, Gaussian Naive Bayes, and residual multilayer perceptron model. The model outputs are then ensembled using a weighted decision fusion technique while being supported by real-time blacklist confirmation and structural risk rules. On the other hand, SMS and email phishing detection methods use text preprocessing steps such as tokenization, stop-word removal, stemming, and TF-IDF vectorization, before the texts are classified with Support Vector Machines. Performing phishing detection purely from URLs, the system has reached the accuracy of 96.28% while having a good balance with other evaluation metrics as well. As well, PhishGuard further has a two-level explanation component that returns both feature-level and natural-language explanations to assist interpretation. The discussed framework is a multi-channel phishing detection system that is not just scalable but also personal to the user

Keywords - Two-Sided Detection, Explainable AI, Ensemble Learning, URL Feature Engineering, Support Vector Machine, Cybersecurity