AIPhiShield: Client-Side Machine Learning For Real-Time Phishing URL And QR Code Threat Detection

Phishing attacks and malicious QR codes constitute two of the most prevalent vectors of cybercrime, accounting for billions of dollars in financial losses annually. Existing defences rely on server-dependent machine learning pipelines or easily bypassed keyword heuristics that produce unacceptable false-positive rates on legitimate sites. This paper presents AIPhiShield, a browser-native cybersecurity tool that replaces heuristic match-ing with a Logistic Regression classifier trained on 20 structural URL features using Python and scikit-learn, then exported as a 2.5 KB JSON weight file and executed entirely within the browser via a custom JavaScript inference engine. No URL is transmitted to any external server for machine learning scor-ing, preserving user privacy. Detection is augmented by cross-referencing the OpenPhish live phishing feed and a curated 52-entry compound-phrase blacklist. The integrated system ad-ditionally provides QR image scanning, live webcam QR scan-ning, an LLM-powered cybersecurity chatbot routed through a Flask proxy that conceals the API key from frontend code, voice input, and geolocation-enriched scan history. The trained model achieves 100% accuracy, precision, recall, and F1-score on a stratified 72-sample test set, with zero false positives and zero false negatives. Feature importance analysis identifies HTTPS usage, high-risk top-level domain, and raw IP address as the three strongest predictors.