👩‍🔬 Authors & Contributors

• Dr. Seyedmostafa Safavi
• Chee Zhung Ong
• Sze Shun Low
• Wenn Yeow Russell Chan
• Zhi Hong Yip

---

🧠 A Lightweight Random Forest Model for Intrusion Detection using UNSW-NB15

This project presents a lightweight Random Forest (RF) model for network intrusion detection, designed to balance high detection accuracy with low computational cost. The model performs efficiently even in real-time and resource-constrained environments, making it suitable for modern cybersecurity systems.

DOI: [XXXXXXX]

---

⚙️ Workflow Overview

1. Load and Prepare Data

• Loads datasets:
  • UNSW_NB15_training-set.csv
  • UNSW_NB15_testing-set.csv
• Removes unnecessary columns: id, attack_cat
• Encodes labels:
  • 0 → Normal
  • 1 → Attack

---

2. Encode and Scale Features

• Categorical features encoded using LabelEncoder
• Numerical features standardized using StandardScaler

---

3. Handle Class Imbalance

• Applies SMOTE (Synthetic Minority Over-sampling Technique)
  to balance Normal and Attack samples.

---

4. Model Training and Hyperparameter Tuning

• Uses GridSearchCV for optimal hyperparameter selection
• Trains each model on the SMOTE-balanced dataset
• Includes baseline and tuned models for comparison

---

5. Evaluation Metrics

Each model is evaluated using:

• Accuracy
• Precision
• Recall
• F1-Score

✅ Confusion Matrices are plotted for visual comparison.
✅ A combined bar chart summarizes all test metrics.

---

6. Model Saving

• Trained models saved in the /models directory as .pkl files
• Encoders and scalers stored for easy reuse during deployment

---

7. Visualization

• Generates visual comparisons of model performance
• Includes confusion matrices and bar charts for clarity

---

🚀 Quick Start

1. Clone the Repository

git clone https://github.com/hadeswithming23/UNSWDLI.git
cd UNSWDLI

2. Install Dependencies

pip install -r requirements.txt

3. Run the Streamlit App

streamlit run unsw_app.py

Then open your browser and go to: http://localhost:8501