🛡️ Global Attack Detection Fusion Model

A unified Fusion Classifier built on top of specialized models for Malware, Phishing, DDoS, and IoT Network Attacks. This global model leverages ensemble learning to enhance cyber threat detection accuracy across varied attack surfaces, with visual interfaces, interactive simulations, and evaluation metrics.

---

📌 Overview

In the current threat landscape, relying on a single model for intrusion detection is insufficient. This project introduces a Global Fusion Model that consolidates predictions from four attack-specific classifiers:

• Malware Detection Model
• Phishing Detection Model
• DDoS Attack Classifier
• IoT Network Intrusion Model

The fusion classifier acts as a decision-maker, improving generalization across datasets, especially with casual/random sampling, noisy inputs, or real-time packet streams.

---

🧠 Attack-Specific Models

Model	Technique
Malware Detector	Graph Sage + OOD Testing
Phishing Classifier	Graph Sage + Casual sampling
DDoS Detector	Graph Sage + Casual sampling
IoT Intrusion Model	Graph Sage + Casual sampling

Each model is trained and validated individually before being integrated into the fusion classifier.

---

🧪 Global Fusion Model

Architecture:
The fusion classifier is built using a meta-learning approach, primarily:

• Soft Voting
• Weighted Averaging
• Stack Ensemble

Features:

• Attack-type-specific preprocessing
• Modular pipelines
• Model confidence tracking
• Automatic fallback to high-confidence local models

---

🎛️ Visual Interface

The dashboard provides:

• Class probabilites of the attack predicted by model.
• Real-time threat classification
• Attack-type probability visualization
• Logs & system warnings
• Accuracy metric graphs
• Probable Reasoning to support gooabl and local prediction.

Built using Streamlit for simplicity and interaction.

---

🧪 Simulation & Causal Sampling

We simulate network behavior under random and causal sampling to:

• Test model generalizability
• Observe misclassification under adversarial inputs
• Measure performance degradation

**Simulation **:

• Visual Interface to demonstate the difference between casual and random sampling with respect to false positive rates and accuracy.

---

📊 Results

Metric	Global Fusion Model
Accuracy	88.00%
Precision	85.29%
Recall	96.67%
F1 Score	90.62%
ROC-AUC	96.25%

---

🚀 Installation

git clone https://github.com/spk-22/Graph-Shield.git
python global.py

For visual inteface

python web_app.py

For simulation demonstrating difference between casual and random sampling, run the below HTML file for visualization

sampling_comparison_simulation.html