multimodal

README.md

@@ -1,128 +1,131 @@
 ---
 license: apache-2.0
-title: EmailGuard
+title: EmailGuard2
 sdk: gradio
-emoji: ⚡
-colorFrom: yellow
-colorTo: purple
+emoji: 🌍
+colorFrom: blue
+colorTo: pink
 short_description: The only secure and rational email phishing detector
 ---
+# EmailGuard2 : Advanced Phishing Detection System
 
-# EmailGuard:  AI-Powered Phishing Detection System
+A multi-model ensemble system for detecting phishing attempts in emails, URLs, and text messages using AI and feature engineering.
 
-The only secure and rational email phishing detector using advanced DistilBERT architecture for multilabel classification of emails and URLs.
+## Features
 
-## Model Architecture
+- Multi-model ensemble prediction
+- Advanced feature extraction and analysis
+- Real-time phishing detection
+- Web-based user interface
+- Risk scoring and confidence reporting
+- URL and email content analysis
 
-**Base Model:** DistilBERT (Distilled Bidirectional Encoder Representations from Transformers)
-- **Task Type:** Multilabel sequence classification
-- **Framework:** Hugging Face Transformers
-- **Fine-tuning:** 3 epochs using Trainer API
-- **Input Length:** Maximum 512 tokens with truncation
-- **Output Classes:** 4-class multilabel classification
+## Installation
 
-## Performance Metrics
-
-- **Accuracy:** 99.58%
-- **F1-Score:** 99.579
-- **Precision:** 99.583  
-- **Recall:** 99.58%
+1. Clone the repository:
+```bash
+git clone <repository-url>
+cd emailguard-phishing-detection
+```
 
-## Dataset
+2. Install dependencies:
+```bash
+pip install -r requirements.txt
+```
 
-Trained on custom dataset `cybersectony/PhishingEmailDetectionv2.0` containing labeled emails and URLs classified as legitimate or phishing attempts.
+3. Run the application:
+```bash
+python app.py
+```
 
-## Classification Categories
+4. Open your browser and go to `http://localhost:7860`
 
-1. **Legitimate Email** - Normal email communications
-2. **Phishing URL** - Malicious web links
-3. **Legitimate URL** - Safe web links
-4. **Phishing Email** - Fraudulent email attempts
+## Usage
 
-## Technical Implementation
+1. Enter email content, URL, or suspicious text in the input field
+2. Click "Advanced Analysis" to process the input
+3. Review the results including risk level and confidence scores
 
-The model uses softmax activation for probability distribution across classes, with the highest probability determining the primary classification. Input preprocessing includes tokenization with padding and truncation to maintain consistent input dimensions.
+## Models Used
 
-## 🚀 Getting Started
+- Primary: `cybersectony/phishing-email-detection-distilbert_v2.4.1`
+- URL Specialist: Custom URL analysis model
+- Feature Engine: Hand-crafted pattern detection rules
 
-### Option 1: Use Online (Recommended)
-**Try EmailGuard instantly - no installation required!**
-1. Visit our live demo on Hugging Face Spaces
-2. Paste your email content or suspicious URL
-3. Click "Analyze for Phishing" 
-4. Get instant results with confidence scores
+## Detection Features
 
-### Option 2: Local Installation
-```bash
-# Clone the repository
-git clone https://huggingface.co/spaces/[your-username]/EmailGuard
-cd EmailGuard
+### URL Analysis
+- Suspicious domain detection
+- Shortened URL identification
+- Malicious link patterns
 
-# Install dependencies
-pip install gradio==5.0.1 transformers torch
+### Content Analysis
+- Urgency keyword detection
+- Money-related terms
+- Personal information requests
+- Spelling error patterns
+- Excessive capitalization
 
-# Run locally
-python app.py
-```
+### Risk Assessment
+- HIGH RISK: Strong phishing indicators (>60% confidence)
+- MEDIUM RISK: Suspicious patterns (30-60% confidence)
+- LOW RISK: Appears legitimate (<30% confidence)
 
-## 💡 How to Use EmailGuard
+## System Requirements
 
-1. **Input:** Paste suspicious email content, URLs, or text messages
-2. **Analyze:** Click the analyze button or press Enter
-3. **Review:** Check the risk assessment and confidence breakdown
-4. **Verify:** Always cross-check results through official channels
+- Python 3.8+
+- 4GB+ RAM
+- Internet connection (for initial model download)
 
-### Example Inputs to Test:
-- Suspicious payment verification emails
-- Unknown links from social media
-- Urgent account security messages
-- Prize/lottery notification emails
+## Technical Details
 
-## 📋 Suggestions & Best Practices
+The system uses:
+- PyTorch for deep learning models
+- Transformers for NLP processing
+- Gradio for web interface
+- Custom ensemble voting mechanism
+- Feature-based risk adjustment
 
-**✅ Good Use Cases:**
-- Educational cybersecurity training
-- Academic research projects
-- Initial screening of suspicious content
-- Learning about phishing patterns
+## Example Inputs
 
-**⚠️ Important Limitations:**
-- This is a prototype for academic purposes
-- Not intended for production security systems
-- Always verify through official channels
-- Combine with human judgment and expertise
-
-## 🤝 Contact & Support
-
-**Questions? Feedback? Collaboration?**
-
-📧 **Email:** [email protected]
+**Phishing Example:**
+```
+URGENT: Your PayPal account has been limited! Verify immediately at http://paypal-security-check.suspicious.com/verify
+```
 
-We welcome:
-- Academic collaboration inquiries
-- Technical feedback and suggestions
-- Bug reports and improvement ideas
-- Research partnership opportunities
+**Legitimate Example:**
+```
+Hi Sarah, Thanks for the quarterly report. Let's discuss in tomorrow's meeting. Best, Mike
+```
 
-## 🎯 Take Action Now!
+## Configuration
 
-**Ready to test EmailGuard?**
-1. **[Try the Live Demo →]** Start analyzing suspicious emails instantly
-2. **[Fork on GitHub →]** Contribute to the open-source project
-3. **[Share with Friends →]** Help others stay safe from phishing
+Model configuration in `app.py`:
+```python
+MODELS = {
+    "primary": "cybersectony/phishing-email-detection-distilbert_v2.4.1",
+    "url_specialist": "cybersectony/phishing-email-detection-distilbert_v2.4.1"
+}
+```
 
-**Stay Safe Online!** 🛡️
+## Limitations
 
----
+- This is an educational/research tool
+- Always verify suspicious content through official channels
+- May produce false positives/negatives
+- Requires manual verification for critical decisions
 
-### Academic Disclaimer
+## License
 
-**Date:** May 30, 2025
+Apache2.0 License
 
-This application is developed as an academic project by University of Dar es Salaam students: _**Byabato, Emmaculata, Regina, Sandy, Gladness, Alvin, Dorcas, and Albert**_.
+## Contributing
 
-**Important Notice:** This tool is intended solely for educational and research purposes. The developers hold no rights, benefits, or responsibilities regarding its use. Users are strongly advised to exercise caution and not rely on this system as a direct security solution. This is a prototype for academic evaluation and should not replace professional cybersecurity tools or expert judgment. Always verify suspicious content through official channels and established security protocols.
+1. Fork the repository
+2. Create a feature branch
+3. Make your changes
+4. Submit a pull request
 
----
+## Support
 
-*Built with ❤️ by University of Dar es Salaam's Computer Science and Engineering (CSE) students*
+For issues and questions, please use the GitHub issue tracker.