Loading…
Preparing your workspace
Understand the three-layer phishing detection pipeline powering PhishGuard.
Every submission passes through three independent analysis layers, then a trained XGBoost machine-learning model combines URL structural features and OSINT signals to produce a phishing probability. NLP text analysis supplements the ML score for email and free-text inputs.
Natural Language Processing with spaCy analyses the textual content for known phishing indicators. The model detects six categories of manipulative language commonly found in phishing emails and messages.
URLs are decomposed into structural features and compared against known attack patterns. This layer catches obfuscation techniques that text analysis alone would miss.
Open Source Intelligence queries external threat-intelligence services to gather contextual data about the domain and hosting infrastructure. This layer is the most expensive but provides strong ground-truth signals.
The system uses an XGBoost gradient-boosted decision tree as its primary classifier. The model was trained on 23,374 labelled samples and evaluated on a held-out test set of 5,009 samples.
The final score maps to one of four threat levels. Each level includes a recommended action for the user.
Submit a URL, email, or text sample and see the pipeline in action.
Start Analysing