Explainable AI-based Phishing Detection using TFIDF- SVD Embeddings and XGBoost Classification

The AI-powered phishing detection system in this work, PhishSentinel, is designed to enhance cybersecurity by accurately identifying malicious emails and URLs. In this approach, a hybrid machine learning model is developed fusing semantic analysis using TF-IDF and SVD embeddings of text with handcrafted URL and text features; an XGBoost model conducts classification for precise detection. A web application based on Streamlit allows real-time analysis, model retraining, and visual explanation of predictions as SHAP-style feature importance charts. This approach improves not only the accuracy of detection but also interpretability to help users understand why an email or URL has been flagged. Contributing to safer digital environments, PhishSentinel directly supports SDG 9: Industry, Innovation, and Infrastructure by promoting secure and resilient digital infrastructure using innovative AI solutions. The proposed approach provides a scalable, explainable, and easy-to-use solution for mitigating phishing attacks in modern communication systems.