**Solving a Temporal Graph Neural Network (TGNN) Challenge f

Source: Dev.to

Scenario

Imagine you are tasked with developing a temporal graph neural network to predict the traffic congestion level in a city’s road network over the next hour, given the current real‑time traffic data, weather conditions, and time of day. The model must incorporate both spatial and temporal graph structures and account for periodic events such as rush hour, festivals, and construction roadblocks.

Constraints

• Graph Size: 1,000 nodes (intersections) with an average of 200 edges, resulting in a dense graph with ~200,000 edges.
• Temporal Resolution: 1‑minute resolution traffic data for the past 24 hours will be used for training.
• Weather Data: Real‑time temperature, humidity, wind speed, and precipitation must be incorporated.
• Model Evaluation: Use a combination of Mean Absolute Error (MAE), Mean Squared Error (MSE), and Area Under the Receiver Operating Characteristic Curve (AUROC).
• Computational Limitations: Training is limited to a single NVIDIA Tesla V100 GPU (16 GB memory) with a maximum of 4 hours of training time.

Objective

Develop a temporal graph neural network that can accurately predict the traffic congestion level at each intersection in the road network over the next hour, given the real‑time traffic data and weather conditions.

Submission Requirements

1. Codebase – A well‑documented Python project using TensorFlow or PyTorch.
2. Model Description – A detailed explanation of the architecture, highlighting any novel graph neural network operations or techniques employed.
3. Performance Plot – Visualizations of the model’s MAE, MSE, and AUROC on the evaluation set.

Evaluation Criteria

• Prediction Accuracy – Quality of traffic congestion forecasts.
• Interpretability – Ability to explain model decisions.
• Computational Efficiency – Effective use of the given hardware and training time.

Submission Deadline

December 15th, 2025.