This research introduces an Advanced Phishing Detection System integrating feature extraction, EDA (Exploratory Data Analysis), and machine learning for real-time threat detection. It evaluates four AI models—Random Forest, SVM, XGBoost, and MLP—achieving high accuracy in identifying phishing websites. Traditional rule-based methods struggle with zero-day attacks, while ML models like Decision Trees and Logistic Regression fail to handle complex patterns effectively. Using a publicly available phishing dataset, the system extracts key features like URL length, domain age, and HTTPS presence. MLP and XGBoost achieve the highest accuracy (99.03% and 95.73% for MLP). A Flask-based web app enables real-time detection. Future work includes LSTMs for sequential URL analysis and continuous learning for adaptive phishing defense. The novelty of this work lies in combining deep learning with real-time deployment via Flask, using a diverse set of phishing indicators beyond basic URL features.