Smart-Shield: A Hybrid ML-AI Framework for Advanced Phishing Detection using Gradient Boosting and LLM Analysis

Quick changes in phishing attempts make sophisticated detection techniques absolutely vital. This work presents a phishing detection system aiming at increasing accuracy and adaptability by combining machine learning with artificial intelligence-driven analysis. Along with Google’s Gemini 2.0 Flash LLM for more in-depth contextual analysis, it makes use of a Gradient Boosting Classifier (GBC) taught on 30 URL characteristics. SMote for class balance and tuned hyperparameters help to improve the model by means of fine-tuning of the detection performance. Two parts characterize the development process: the first phase concentrates on establishing the ML model, which comprises URL categorization, shortlink identification, and phishing pattern analysis; the second phase employs JSP and NetBeans to build a Java-based web application. Three portals comprise the system: an administrative portal for user management, phishing trend analysis and URL classification; a user portal for secure web browsing, suggestions, and automatic phishing alerts; and an attacker simulation site to test phishing detection capability. Python is connected with NetBeans to provide real-time classification and security analysis; an automatic access control method warns or temporarily limits users who surpass limits on browsing harmful URLs. This research presents a very accurate phishing detection system with real-time alarms and enhanced protection against changing threats by combining machine learning with AI-driven analysis.