Evaluating and Enhancing Adversarial Robustness of ML Models for NIDS (CIC-IDS2017)

Applicant: Hitvika Pathak
Programme: IIT Kanpur B.Cyber
Repository: https://github.com/Hitvikapathak/adversarial-nids-cicids2017

Project Goal

Analyze vulnerability of ML-based network intrusion detection to adversarial attacks and implement defenses with practical security relevance.

Repository Structure

adversarial-nids-cicids2017/
├── data/
│   ├── raw/                 # CIC-IDS2017 CSV (auto-downloaded)
│   └── processed/           # processed splits + profile
├── src/
│   ├── config.py
│   ├── data.py
│   ├── models.py
│   ├── attacks.py
│   ├── defenses.py
│   ├── evaluation.py
│   ├── visualization.py
│   └── pipeline.py
├── notebooks/
├── results/
│   ├── screenshots/
│   └── *.png, *.csv, metrics.json
├── docs/
│   ├── personal_reflection.md
│   └── demo_video_script.md
├── scripts/
│   ├── generate_report.py
│   └── generate_one_page_summary.py
├── run.py
└── requirements.txt

Quick Start

pip install -r requirements.txt
python run.py
python scripts/generate_report.py
python scripts/generate_one_page_summary.py

Reproducibility

• Seed: 42
• Split: 72/8/20 train/val/test (stratified)
• Primary epsilon: 0.05 (L∞)
• PGD: 20 steps, step size 0.01

Key Findings (from latest run)

See results/metrics.json and results/summary_table.csv.

Deliverables

1. Main report: Adversarial_Robustness_CIC_IDS2017_Project_Report_FINAL.docx
2. One-page summary: Adversarial_NIDS_One_Page_Summary_IITK_BCyber.docx
3. Figures: confusion matrices + epsilon curve in results/
4. Terminal screenshot: results/screenshots/terminal_output.png
5. Personal reflection: docs/personal_reflection.md
6. Demo script: docs/demo_video_script.md

Project Status

Component	Status
Code + pipeline	Done
Experiments + metrics	Done
Report (17 sections)	Done
One-page summary	Done
Screenshots + graphs	Done
Git commit (local)	Done
GitHub push	Done — https://github.com/Hitvikapathak/adversarial-nids-cicids2017
Demo video	Optional — see docs/DEMO_RECORDING_GUIDE.md

License

Academic portfolio project for IIT Kanpur B.Cyber application.