🛡️ PhishGuard AI

Advanced AI-powered email security for phishing detection

🚀 Overview

PhishGuard AI is a cutting-edge phishing detection system that combines advanced machine learning techniques with psychological intent analysis to provide superior email security. Unlike traditional keyword-based filters, PhishGuard AI analyzes the psychological manipulation techniques used in phishing attacks, making it effective against novel threats.

✨ Key Features

• 🧠 Intent-Based Detection: Analyzes psychological manipulation techniques beyond simple keywords
• 🔗 Ensemble Learning: Combines multiple ML models (Naive Bayes, Random Forest, SVM) with soft voting
• 📊 Advanced Features: TF-IDF with n-grams (1,2,3,4) + psycholinguistic analysis
• ⚡ Real-time Analysis: Instant email security assessment with confidence scores
• 🎨 Modern UI: Beautiful, responsive web interface with detailed result visualization
• 🔒 Privacy-First: Email content processed locally, not stored

🎯 Performance

• Test Accuracy: 96.77%
• Training Samples: 3,865 emails across 3 classes
• False Negative Reduction: Improved from 0.26 to 0.94+ confidence on critical cases
• Intent Signal Detection: 15+ psychological manipulation features

🏗️ Architecture

Machine Learning Pipeline

Email Text → TF-IDF Vectorization → Intent Feature Extraction → Ensemble Model → Prediction + Confidence

Dataset Classes

• AI-Generated Phishing: Machine-generated phishing attempts
• Human-Written Phishing: Real-world phishing campaigns
• Legitimate Emails: Normal business communications

Intent Analysis Features

• Authority Signals: Official/government language detection
• Pressure Tactics: Urgency and deadline analysis
• Scarcity Signals: Limited time offer detection
• URL Analysis: Suspicious link identification
• Readability Metrics: Text complexity analysis

🛠️ Technology Stack

• Backend: Python, Flask, Scikit-learn
• Frontend: HTML5, CSS3, JavaScript, Bootstrap 5
• ML Libraries: Pandas, NumPy, SciPy
• Features: TF-IDF, N-grams, Ensemble Methods

📦 Installation

Prerequisites

• Python 3.8 or higher
• pip package manager

Quick Start

1. Clone the repository

   git clone https://github.com/EclipseAditya/phishguard-ai.git
   cd phishguard-ai

2. Install dependencies

   pip install -r requirements.txt

3. Ensure model files exist Make sure these files are in the models_and_dataset/ folder:

   • Advanced_enhanced_model.pkl
   • enhanced_tfidf_vectorizer.pkl

4. Run the application

   python app.py

5. Open your browser Navigate to http://localhost:5000

💻 Usage

Web Interface

1. Open PhishGuard AI in your browser
2. Paste email content into the analysis textarea
3. Click "Analyze Email Security"
4. View detailed results including:
   • Classification (AI Phishing / Human Phishing / Legitimate)
   • Confidence scores for each class
   • Intent analysis with manipulation signals
   • Risk level assessment

API Endpoint

POST /api/analyze
Content-Type: application/json

{
  "email_text": "Your email content here..."
}

Response:

{
  "prediction": "ai_phishing",
  "confidence_scores": {
    "ai_phishing": 0.89,
    "human_phishing": 0.08,
    "legitimate": 0.03
  },
  "intent_analysis": {
    "authority_signals": 2,
    "pressure_signals": 3,
    "scarcity_signals": 1,
    "total_manipulation_signals": 6,
    "readability_score": 65.2,
    "urls_detected": true
  },
  "risk_level": "HIGH",
  "risk_color": "danger"
}

🔬 Technical Deep Dive

Feature Engineering

• TF-IDF Vectorization: Captures word importance with n-gram context
• Psycholinguistic Features: Authority, pressure, scarcity signal detection
• URL Analysis: Link structure and suspicious domain detection
• Readability Metrics: Text complexity and manipulation indicators

Model Architecture

• Base Models: MultinomialNB, RandomForest, LinearSVM
• Ensemble Method: Soft voting for probability averaging
• Cross-Validation: Stratified K-fold for robust evaluation
• Feature Combination: TF-IDF + Intent features via sparse matrix concatenation

Security Focus

PhishGuard AI prioritizes False Negative reduction over raw accuracy, as missing a phishing email poses greater security risk than flagging a legitimate email.

📊 Results Visualization

The web interface provides:

• Risk Level Badges: Visual risk assessment (HIGH/LOW)
• Confidence Bars: Animated confidence scores for each class
• Intent Signals: Color-coded manipulation technique detection
• Warning Alerts: Highlighted threats with signal counts

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

👨‍💻 Developer

Aditya Pandey
AI/ML Developer & Security Researcher

Passionate about applying artificial intelligence to cybersecurity challenges. Specialized in machine learning, natural language processing, and threat detection systems.

🙏 Acknowledgments

• Advanced machine learning techniques for email security
• Modern web development practices for user experience
• Cybersecurity research community for threat intelligence
• Open source ML libraries that made this possible

---

PhishGuard AI - Protecting your digital communications with advanced artificial intelligence