---
license: mit
language:
- en
metrics:
- accuracy
library_name: keras
tags:
- cybersecurity
- phishing-detection
- qr-code-analysis
---

# AEGIS-SWARM: Visual Phishing Auditor
**Developer:** Muhammad Abdullah  
**Institution:** COMSATS University Islamabad, Lahore Campus

## Overview
This model is a **Convolutional Neural Network (CNN)** developed as part of the AEGIS-SWARM multi-modal threat triage system. It is specifically designed to analyze images (such as QR codes) to determine if they lead to malicious phishing sites.

## Technical Specifications
- **Architecture:** Sequential CNN (Conv2D, MaxPooling, Dense layers)
- **Input Shape:** (128, 128, 3)
- **Framework:** TensorFlow/Keras

## Training Progress
The model was trained on the **CIC-Trap4Phish** dataset, involving over **1.5 million images**.
- **Epochs:** 5
- **Final Accuracy:** 63.61%
- **Final Loss:** 0.5798

## Usage
To use this model in your own Python environment:
```python
from tensorflow.keras.models import load_model
from huggingface_hub import hf_hub_download

# 1. Download the weights
model_path = hf_hub_download(repo_id="Wa1106/AEGIS-SWARM-Visual-Agent", filename="visual_agent_v1.h5")

# 2. Load the model
model = load_model(model_path)

# 3. Predict
# results = model.predict(your_preprocessed_image)
\```
---