Cities run on webs of roads, pipes, and routes. When something breaks, the shock spreads fast. A new study dives into how we can forecast those breakdowns before they cause real damage.
Researchers built a three-step system. First, they mapped how the network is connected. Then, they used a tool called a Graph Convolutional Network, or GCN, to study traffic and flow patterns. Finally, they stress tested the whole setup to see what happens when parts of the network go offline. The team used a giant dataset from New York City, covering over 6,000 infrastructure points and nearly 97,000 disruption events.
The GCN model did impressively well. It beat older methods like ARIMA, LSTM, and XGBoost by a wide margin. The model hit an error rate of just 5. 8 and scored 0. 92 on accuracy. On top of th