Increasing the Business Value of Free-Floating Carsharing Fleets by Applying Machine-Learning Based Relocations

Free-floating carsharing (CS) services provide customers with a fleet of vehicles distributed within an operation area. These services gained popularity because of their positive impact on societal and personal mobility. Understanding determinants of customer demand is a key challenge for developing and applying vehicle relocation strategies to prevent the formation of undersupply areas. In this study, we merge possible features from publicly available data sources with historical demand from CS services situated in three different-sized cities. We train and test a Random Forest (RF) regressor estimating demand based on the enhanced dataset. Building on this demand prediction, we developed a relocation strategy that optimizes vehicle availability at anticipated demand points. Our strategy improved the reservation acceptance ratio in all three reference systems between 7.1 % and 15.6 %. Furthermore, the number of relocations compared to a deterministic relocation strategy could be reduced by 82.3 % and 20.6 % in two cities.