AI-Driven Attack Defense for IoT Smart Cities in 6G Networks

Overview

This project was developed as part of the PIDATA – 4th Year Engineering Program at Esprit School of Engineering – Tunisia (Academic Year 2025–2026).

The objective of this project is to design an AI-driven framework for detecting and mitigating cyber-attacks in IoT-based smart city environments within emerging 6G networks. The system leverages data analytics and machine learning techniques to identify anomalous network behavior and enhance network security.

---

Problem Statement

With the rapid growth of IoT devices in smart cities and the evolution toward 6G networks, cybersecurity challenges have become more complex. Traditional rule-based systems are no longer sufficient to detect sophisticated and evolving attacks.

This project addresses:

• Detection of malicious network traffic
• Handling heterogeneous 5G/6G datasets (Global, eMBB, mMTC, URLLC)
• Building intelligent models for real-time anomaly detection

---

Features

• ✅ Data preprocessing pipeline (cleaning, missing values, encoding)
• ✅ Outlier handling using IQR Winsorization
• ✅ Feature engineering (log transformations, correlation filtering)
• ✅ Categorical encoding (One-Hot Encoding)
• ✅ Multi-dataset analysis (5G traffic scenarios)
• ✅ Machine Learning-based anomaly detection+DL for unlabeled 6G datset
• ✅ Balanced dataset evaluation and performance metrics

---

Tech Stack

Data Processing & Machine Learning

• Python
• Pandas
• NumPy
• Scikit-learn

Data Visualization

• Matplotlib
• Seaborn

Optional Extensions

• Streamlit (interactive dashboard)
• Docker (containerization)
• GitHub Actions (CI/CD)

---

Dataset Description

The project uses multiple datasets representing different 5G traffic scenarios:

• Global: Mixed traffic dataset
• eMBB: Enhanced Mobile Broadband
• mMTC: Massive Machine Type Communications
• URLLC: Ultra-Reliable Low Latency Communications

Each dataset includes:

• Network traffic features (packets, bytes, latency, etc.)
• Protocol information
• Target label: Benign (0) / Malicious (1) Also 6G datset

Dataset Access

Due to file size limitations on GitHub, the datasets used in this project are not included in the repository.

They can be accessed through the following link:

👉 Download Dataset : https://drive.google.com/drive/folders/1QHhjL0muKQa_dtk52HXc6WW7MTUyG1nR?usp=sharing

---

Data Preprocessing Pipeline

The following steps were applied:

1. Removal of irrelevant features (IDs, constant columns)
2. Handling missing values using median imputation
3. Cleaning categorical anomalies (? values)
4. Removing highly correlated features
5. Outlier treatment using IQR-based Winsorization
6. Log transformation for skewed features
7. One-Hot Encoding for categorical variables
8. Final dataset validation (no missing values, aligned schema)

---

Machine Learning Approach

Problem Type

• Binary Classification (Benign vs Malicious)

Models used

• Logistic Regression
• Random Forest
• XGBoost
• Isolation Forest (for anomaly detection scenario)

Evaluation Metrics

• Accuracy
• Precision / Recall
• F1-Score
• ROC-AUC
• Confusion Matrix

---

Results

5G best model XGboost:

• Accuracy: 99%
• F1-Score: 94%
• ROC-AUC: 98%

---

Project Structure

📁 project-root
│
├── 📁 data/
│   ├── Global_clean.csv
│   ├── eMBB_clean.csv
│   ├── mMTC_clean.csv
│   └── URLLC_clean.csv
│
├── 📁 notebooks/
│   ├── data_cleaning.ipynb
│   ├── eda.ipynb
│   ├── modeling.ipynb
│
├── 📁 src/
│   ├── preprocessing.py
│   ├── training.py
│   ├── evaluation.py
│
├── 📁 models/
│   └── model.pkl
│
├── requirements.txt
└── README.md

---

Getting Started

1. Clone the repository

git clone https://github.com/your-username/your-repo.git
cd your-repo

2. Install dependencies

pip install -r requirements.txt

3. Run the project

• Open notebooks using Jupyter or Google Colab
• Execute preprocessing → modeling → evaluation

---

Supervisors

• Bouraoui Rahma
• Cherif Safa
• Mejri Ameni

---

Academic Context

Developed at Esprit School of Engineering – Tunisia PIDEV – 4rd Year Engineering Program Academic Year: 2025–2026

---

Future Work

• Integration with real-time streaming data
• Deployment using Streamlit dashboard
• Deep Learning models (LSTM, Autoencoders)
• MLOps pipeline (CI/CD + monitoring)

---

Acknowledgments

• Esprit School of Engineering
• Academic supervisors and project mentors
• Open-source community