Ride-hailing algorithm cuts wait times by 31% — without retraining on new data

Researchers built a system that matches current traffic patterns to similar days in the past, then uses those historical patterns to position cars and dispatch rides. The result: a ride-hailing company running this algorithm would cut average wait times from, say, 10 minutes to 7 minutes, and do it without constantly retraining the system on new data.

Ride-hailing dispatch is a constant optimization problem — demand shifts by hour, by season, by weather, by whether there's a concert or a snowstorm. Current systems either retrain constantly (expensive) or fail when conditions change. This approach solves that by asking a simpler question: what day in the past looked most like today? If you can answer that, you already know where riders are and where they're going. The practical effect is dramatic: tested on 5.2 million real NYC rides, it cuts wait times by nearly a third, and the improvement holds in Chicago without any retraining. For a company like Uber or Lyft, this is a direct line to less frustrated riders and fewer cars sitting idle.