Trip Inference

Trip Inference is a means to determine what trip a vehicle is on based on various data sources, such as time of day, GPS location and block assignments. Performance for Trip Inference is measured by running against a set of archived real-world trips, and then calculating the percentage of how trip IDs inferred by the system were actually correct. A score of 0% means no inferred IDs were correct, while 100% signifies that all of the inferred IDs were correct.

After initial implementation of the system, we measured a baseline of 59%.

A series of performance optimizations took us to 85%.

And finally, including block assignment data in the inference process gave us a score of 97% for the archived trips.