junzis/atdelay

Implements three complementary prediction models across different scales: Random Forest for individual flight arrival delays, LSTM networks for single-airport aggregate delays, and Dynamic Spatial-Temporal Graph Attention Networks (DST-GAT) for network-wide airport delay forecasting. Uses EUROCONTROL R&D data with TensorFlow/Keras backends, supporting optional GPU acceleration via CUDA for computationally intensive graph neural network training. Integrates the Spektral library for graph convolution operations to capture airport connectivity and temporal delay propagation patterns.

No commits in the last 6 months.