Path complexity for observed and predicted bicyclist routes

Abstract

Everyday route choices made by bicyclists are known to be more difficult to explain than vehicle routes, yet prediction of these choices is essential for guiding infrastructural investment in safe cycling. In this paper we study how the concept of route complexity can help generate and analyze plausible choice sets in the demand modeling process. % Defining route complexity to be the minimal number of shortest path segments that form a given complete route, we characterize the routes bicyclists take in large set of GPS traces gathered voluntarily by persons traveling to everyday activities at work, school, friends, etc. The complexity of a given path in a graph is the minimum number of shortest paths that is required to specify that path. Complexity is a path attribute which is considered to be important for route choice in a similar way as the number of left turns, the number of speed bumps, distance and other. The complexity was determined for a large set of observed routes and for routes in the generated choice sets for the corresponding origin-destination pairs. The respective distributions seem to significantly differ so that the choice sets do not reflect the traveler preferences. This paper looks at how the observed routes compare to routes generated by Breadth First Search Link Elimination and Double Stochastic Generation Function method.