The algorithm can compute the likelihood of a vehicle running a red light based on its rate of deceleration as it is approaching the intersection with a level of precision down to mere milliseconds. The team, which applied the algorithm to more than 15,000 vehicles during the study, used instruments that monitored vehicle speeds and locations as well as when the lights turned red. When the results were tallied, they found that they were able to correctly predict who would run a red light 85 percent of the time.