AI Powered Phishing Email Detector with Gmail Live Scanning

Abstract: Phishing attacks remain the most prevalent cyber threat, responsible for over 90% of data breaches and causing billions of dollars in financial losses annually. Traditional rule- based email filters fail to detect sophisticated, AI- generated phishing emails that evade signature-based detection. This paper presents PhishGuard, a novel hybrid dual- stage classification framework that combines Random Forest ensemble learning with a CNN-LSTM deep neural network for real-time phishing email detection. Our system extracts 25 engineered features from email headers and content, including SPF/DKIM authentication status, URL analysis, and linguistic patterns. The framework employs a weighted decision fusion mechanism that adjusts confidence scores based on sender authentication and trusted domain verification. We implement a complete web-based solution with FastAPI backend and React frontend, featuring seamless Gmail API integration via OAuth 2.0 for real-time inbox scanning. Experimental results demonstrate that PhishGuard achieves 96.3% accuracy, 96.1% precision, and 96.5% recall on our evaluation dataset, outperforming single-model approaches by 2.1%. The system processes emails in under 165ms, making it suitable for real-time deployment. Our contribution includes the complete open-source implementation, a comprehensive feature engineering pipeline, and a user-friendly interface that educates users about phishing indicators while protecting them.

Keywords: Phishing Detection, Deep Learning, Random Forest, CNN-LSTM, Email Security, Natural Language Process- ing, Gmail API, Cybersecurity