GCN-LOF: A Graph Convolutional Network with Local Outlier Factor

Overview

The paper 'Investigation of Intrusion Detection Methods for Smart Systems Utilizing the GCN-LOF Algorithm' code. This repository implements GCN-LOF, a method that combines Graph Convolutional Networks (GCNs) with the Local Outlier Factor (LOF) for anomaly detection and other graph-based learning tasks. The code is written in Python and leverages popular libraries such as PyTorch for deep learning.

Features

• Graph Convolutional Networks (GCNs): Leverages GCNs to capture graph structure and node features.
• Local Outlier Factor (LOF): Enhances anomaly detection by integrating LOF with GCN outputs.
• Customizable: Modular design allows easy extension for new datasets and models.

Repository Structure

GCN-LOF-main/
├── gcn-lof-chinese.py        # Example script in Chinese for specific use cases
├── main.py                   # Main entry point for the project
├── data/
│   └── full_dataset/         # Folder containing datasets
├── models/
│   └── gcn.py                # Implementation of the GCN model
├── train/
│   └── train_test.py         # Training and testing scripts
├── utils/
│   ├── data_loader.py        # Data loading utilities
│   ├── metrics.py            # Performance metrics

Installation

Prerequisites

• Python >= 3.8
• PyTorch >= 1.10
• Other dependencies listed in requirements.txt

Steps

1. Clone the repository:

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

2. Install dependencies:

   pip install -r requirements.txt

3. Verify installation by running a test script:

   python main.py --help

Usage

1. Data Preparation

Place your datasets in the data/full_dataset/ directory. Ensure the data is formatted according to the instructions in data/full_dataset/README.md.

2. Training the Model

Run the training script:

python main.py --train --epochs 50 --dataset your_dataset_name

3. Testing the Model

Evaluate the trained model:

python main.py --test --dataset your_dataset_name

4. Hyperparameter Tuning

Modify hyperparameters in the main.py or pass them via command-line arguments, e.g.,:

python main.py --train --learning_rate 0.01 --hidden_units 64

Key Scripts

• main.py: Entry point for training and testing the GCN-LOF model.
• models/gcn.py: Defines the GCN architecture.
• utils/data_loader.py: Handles data preprocessing and loading.
• utils/metrics.py: Implements performance metrics.

Example

Below is an example to train and evaluate the model:

python main.py --train --dataset cora --epochs 100
python main.py --test --dataset cora