A smart transportation system can be seen as an aggregate of transportation opportunities and services, accompanied by advanced management services that make the access to the system easier for the user. In this paper, we exploit the product line paradigm to address the variability of an exemplary smart transportation system: a bike-sharing system. Improving the satisfaction of a user of a bike-sharing system includes providing information at runtime on the filling degree of the docking stations in the near future. To fulfill this expectation, a prediction service is needed to infer the probability that at a certain time of the day a user will return a bike to or take one from a station. In earlier studies, several possible advanced smart predictive services were identified. The choice of which services to offer to users by the managers of a bike-sharing system is influenced by minimizing the costs while maximizing customer satisfaction. To aid the managers, we modeled a family of smart bike-sharing services, after which an attributed feature model was used to augment the model with quantitative attributes related to cost and customer satisfaction, allowing for a multi-objective optimization by dedicated tools. We observe that the performance of the smart prediction services, and therefore of the related customer satisfaction, is highly dependent on the amount of collected historical data on which the predictive analysis is based. Therefore the result of the optimization also depends on this factor, which evolves over time.