AIS Graph Classification

Graph Neural Networks for Maritime Vessel Activity Classification

This tutorial demonstrates how to use Graph Neural Networks (GNNs) to classify vessel fishing activities from Automatic Identification System (AIS) time-series data, using the Deep Graph Library (DGL).

Key Result: GraphSAGE achieved 94.4% test accuracy on the fishing vs. non-fishing classification task.

Episodes

• Introduction
  • Overview
  • Background
  • Why Graphs?
  • Architecture Overview
  • Three GNN Architectures
  • Using AI Coding Assistants
  • Expected Results at a Glance
  • Self-Contained Repository
  • What You Will Learn
• Provisioning a VM
  • NAIC Orchestrator
• Setting Up the Environment
  • Quick Start
  • What setup.sh Does
  • Manual Installation
  • Verify Installation
  • Jupyter Kernel Setup
  • Data Requirements
• Graph Neural Network Theory
  • From Time-Series to Graphs
  • Why Message Passing Works for Trajectories
  • The Message Passing Framework
  • Three GNN Architectures
  • Graph Classification Pipeline
  • Over-Smoothing and Depth
• AIS Data and Graph Construction
  • Automatic Identification System (AIS)
  • Feature Engineering
  • Graph Construction Code
  • Dataset Statistics
• Running the Demonstrator
  • Option A: Jupyter Notebook
  • Option B: Command Line
  • Understanding the Results
  • Parameter Tuning
  • Bootstrap Indices
  • Background Training
• FAQ and Troubleshooting
  • Frequently Asked Questions
  • Common Issues
  • Performance Tips
  • Getting Help

Resources

• Downloads