Maritime Supply Chain Port Congestion Predictor with Graph Neural Networks and PyTorch Geometric
El Niño climate patterns disrupt major global shipping choke points: severe droughts reduce vessel drafts through the Panama Canal, while severe typhoons force port closures across East Asia. Supply chain business interruption insurers require predictive network models capable of mapping risk propagation across global maritime transit graphs. In this challenge, you will construct a dynamic Graph Neural Network (GNN) using PyTorch Geometric (PyG). The model will represent ports as nodes and shipping lanes as edges, ingesting weather severity indicators and vessel AIS tracking density. Your solution will predict port delay hours, estimate supply chain business interruption losses, and identify bottleneck nodes across the global maritime network.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a Graph Neural Network with PyTorch Geometric to predict maritime port congestion and supply chain disruption during El Niño weather events.
How work is evaluated
Evaluates GNN prediction precision for port dwell time delays across global maritime nodes.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
graph_connectivity_check
GNN input adjacency matrix forms connected non-empty graph
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
port_delay_mae
Mean Absolute Error on predicted port delay hours • target: 4.5 • range: 0-50
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Construct dynamic spatio-temporal graph representations using PyTorch Geometric.
Incorporate node attributes (port capacity, weather index) and edge features (lane distance, vessel traffic).
Train Graph Convolutional Networks (GCN / GAT) for node regression tasks.
Predict port congestion dwell times and financial delay risks across shipping networks.
Reference links and supporting material
Synthetic maritime shipping network dataset comprising port node coordinates, daily vessel counts, regional weather severity scores, and lane distances.
Challenge setup
Evaluates GNN prediction precision for port dwell time delays across global maritime nodes.
Challenge input
JSON with port_unlocode, incoming_vessel_count, regional_storm_severity
PyTorch Geometric
Standard deep learning library for graph neural network models.
Evaluated output
JSON with predicted_delay_hours, congestion_level, risk_propagation_index
- GNN input adjacency matrix forms connected non-empty graph
- Mean Absolute Error on predicted port delay hours • target: 4.5 • range: 0-50
- Benchmark: MaritimeSupplyBench
- Port Delay Mae target: 4.5
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- fal.ai · Optional
- Pieces for Developers · Optional
- DeepSeek V3 · Required
Evaluation contract
- graph_connectivity_check · Weight 1
- port_delay_mae · Weight 1
Metrics
- port_delay_mae · Target 4.5
Benchmarks
- MaritimeSupplyBench
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with port_unlocode, incoming_vessel_count, regional_storm_severity
Output: JSON with predicted_delay_hours, congestion_level, risk_propagation_index
Execution harness: Python
Evaluation tools
- PyTorch Geometric · GNN Library
Recipe freeze marker
No recipe freeze time is recorded.