🔐 Web Attack Detection Using Deep Learning (CNN + BiLSTM )

This project explores two different deep learning approaches for detecting web-based cyber attacks using network traffic features. It builds upon prior work, then introduces a creative, dual-path semantic-enhanced model that shows significant performance improvements.

---

📁 Project Structure

project-root/
│
├── notebooks/
│ ├── initial_code.ipynb # Baseline model implementation
│ └── improved_code.ipynb # Improved model with semantic features + attention
│
├── reports/
│ ├── initial_report.md # Report for the baseline model
│ └── improved_report.md # Report for the improved model
│
├── requirements.txt # Python dependencies
├──  README.md # This file
└── result.md #Model Performance Comparison

---

✨ Highlights

• 📊 Dataset: Balanced web attack dataset (Brute Force, SQL Injection, XSS)
• 🧠 Model 1: A basic deep learning model for web attack classification
• 🚀 Model 2: A novel architecture that combines:
  • Semantic feature extraction
  • Convolutional + BiLSTM layers
  • Attention mechanism
  • Dual-input fusion strategy
• 📈 Improved Accuracy, Precision, Recall, and F1-score across 5-fold cross-validation

---

🧪 Datasets

We use the Balanced Web Attacks Dataset originally introduced in this project.

To load it inside your notebook:

!wget https://github.com/fisher85/ml-cybersecurity/blob/master/python-web-attack-detection/datasets/web_attacks_balanced.zip?raw=true -O dataset.zip
!unzip -u dataset.zip

📌 How to Run

1. Clone this repository:

git clone https://github.com/nlppaperclub/web-attack-detection-4033614008.git
cd web-attack-detection-4033614008

2. Install dependencies:

pip install -r requirements.txt

3. Open the notebooks: You can use Jupyter or Google Colab. Each notebook is self-contained and runnable:

notebooks/initial_code.ipynb → Baseline

notebooks/improved_code.ipynb → Enhanced model with attention and semantic logic

🧾 Reports

In-depth explanations and analysis are available in the reports/ folder:

initial_report.md: Describes the baseline model, features, and results

improved_report.md: Outlines the motivation, semantic idea, model improvements, and metrics

🛠️ Tools & Libraries

Python 3.8+

TensorFlow / Keras

scikit-learn

pandas / numpy

matplotlib

⭐ Acknowledgments

Inspired by the original research from fisher85/ml-cybersecurity Semantic feature engineering and architectural improvements proposed independently in this project.