SentinelURL: AI Framework for Predictive Phishing Intelligence

Abstract: Phishing websites copy trusted services to trick people into handing over passwords, banking details and other private data. Blacklists catch links that have already been reported, but they say nothing about a URL that was created or changed an hour ago. This paper presents SentinelURL, a machine-learning framework that judges a URL from its own lexical and structural characteristics before the user opens it. The workflow takes a URL, validates and preprocesses it, turns it into a numerical feature vector, and passes that vector to a trained classifier. The features include URL length, the number of dots and special characters, suspicious keywords, IP-address use, subdomain patterns, entropy and signs of obfuscation or typosquatting. The project report names Random Forest as the core classifier, and the project presentation lists Logistic Regression and XGBoost as models to compare it against. A Python backend and a simple interface return a clear category and warning, so that users can decide more safely. The project materials do not contain a full set of measured results, so this paper reports no accuracy, precision, recall, F1 or latency figures. It sets out the design, the implementation, the evaluation protocol, the present limitations and the steps needed for a rigorous test.

Keywords: phishing URL detection, machine learning, Random Forest, lexical features, cybersecurity, URL classification, predictive threat analysis