DP Network Experiment

This repository contains an experiment using Differential Privacy (DP) with the Opacus library on the UNSW-NB15 network security dataset.

This notebook compares the performance of a baseline model and a differentially private model using PyTorch.

Contents

• Data preprocessing (SMOTE, normalization)
• Model training (Baseline and DP)
• Evaluation (Accuracy, Confusion Matrix)
• Epsilon (ε) calculation using Opacus

Run on Google Colab

Open in Colab

Installing dependencies in Colab

After opening in Colab, run the following command in the first code cell to install required packages:

!pip install torch torchvision opacus scikit-learn imbalanced-learn

Additional Information

The UNSW-NB15 dataset is not included in this repository. Please upload it manually when using Colab.

The notebook includes both baseline and differentially private model evaluation using ε metrics.