Evolution of a Speeding Violation Detection Algorithm Under GPS Spoofing

Maria Medvedeva24 May 2025

Problem

The trip safety team at Yandex Taxi finds drivers who exceed the speed limit and applies penalties. The algorithm works on raw driver coordinates. In 2022 GPS jammers appeared, and spoofed coordinates (the phone is in one place, the location shows another) produced masses of false speeding detections. Honest drivers got undeserved penalties and flooded support with appeals. Spoofing was present in more than 50 cities of the European part of Russia, and the team had to switch the algorithm off city by city.

Example from the article: on maps of detected violations, spoofed points form circles that drift out to sea, where drivers are "riding motorboats".

Solution in brief

The team did not try to detect spoofed points directly. They trust only points that lie within 10 meters of major roads, because prolonged speeding happens on highways and arterials, not inside neighborhoods. This restored the algorithm to 87% of the violations it detected in 2022.

Questions this case answers
  • How to detect rule violations from noisy device location traces when part of the location points are falsified
  • How to filter falsified location points with a reference geometry instead of classifying each point as genuine or fake
  • How to detect speeding by drivers when GPS coordinates are spoofed
  • How to filter spoofed geolocation without a model that classifies each point

Tools and methods

  • Average speed over 30-second intervals — speed from raw coordinates and the road-graph speed limit instead of navigator speed.
  • Road-graph filtering — keep only points close to non-neighborhood roads.
  • Quadkey filtering by share of suspicious orders — a square of about 1 square kilometer is trusted if under 30% of its orders look spoofed.
  • Shortest-path route reconstruction — fills gaps where the driver switched GPS off.
Continue reading — it's free

Google or a one-time email code. No password.