Large sums of money are lost every year, as working hours are spent waiting in traffic on congested roads. To avoid this loss, several methods for travel-time estimation has been developed. By determining the travel-time for a particular piece of road ahead of time, congested roads can be avoided.
Traditionally, etimating travel times has relied on slow and costly methods such as loop detectors, observations vehicles or automatic vehicle identification. Other approaches rely on simple calculations based on road lengths and permitted speeds. These approaches are not able to predict traffic, and therefore only give an estimate that applies outside rush hours. Using consumer products to gather data, however, the process can become faster and cheaper as GPS receivers become more abundant.
In this article we develop two approaches to travel-time estimation, the point-based approach and the trip-based approach. Using data from two different data sources as a starting point, we have developed a solution that is able to use very basic data, while still utilizing additional information. We introduce a data warehouse for storing GPS data, a road network and additional data such as information about drivers and vehicles.
In our experiments, we show how the two approaches perform in general and compared to each other. Using our trip-based approach, we are able to provide travel-time estimates with an error rate of 0.3% compared to the actual travel times, which is a major improvement over naive travel times and a slight improvement over our point-based approach.