Expanding a(n) (electric) bicycle-sharing system to a new city: Prediction of demand with spatial regression and random forests. (April 2020)

• Record Type: Journal Article
• Authors: Guidon, Sergio; Reck, Daniel J.; Axhausen, Kay
• Is Part Of: Journal of transport geography. Volume 84(2020)
• Publication Date: 2020-04
• Subjects: E-bike sharing -- Bicycle sharing -- Demand prediction -- Vehicle sharing -- Spatial regression -- Random forests; Transportation -- Periodicals; Telecommunication -- Periodicals; Transport -- Périodiques; Télécommunications -- Périodiques; Telecommunication; Transportation; Periodicals; 388
• DOI: 10.1016/j.jtrangeo.2020.102692
• Languages: English
• ISSNs: 0966-6923

Abstract

Abstract: Bicycle-sharing systems have experienced strong growth in the last two decades as part of a global trend that started in the 1990s and accelerated after 2005. Early bicycle-sharing systems were provided primarily as a public service by cities. Today, major international bicycle-sharing companies are emerging and seeking to expand their operations to new cities. Two major strategic questions arise: (1) which cities should be considered for expansion and (2) what should be the geographical extent of the service area? An important factor in such decision-making is the expected demand for bicycle-sharing because it relates directly to potential revenue. In this paper, booking data from an electric bicycle-sharing system was used to estimate and assess models for bicycle-sharing demand and to predict expansion to a new city. Employment, population, bars, restaurants and distance to a central location were amongst the most important predictors in terms of variance explained in the same city. Omitting centrality measures improved predictions for the new city. Highlights: Validated bicycle-sharing demand predictions for an expansion to a new city. Model transferability: Omitting centrality measures (such as CBD distance) may improve predictive accuracy for new cities. Model selection: Spatial regression outperformed random forests due to their ability to take spatial dependence into account.