Analyzing the success determinants of on-demand microtransit services and understanding historical user rides provide valuable insights for operators to improve further operation. This study first summarizes the barriers, failure factors, and success determinants of microtransit services and proposes a new methodology based on discrete Bayesian Networks (BNs) to model user’s sequential rides, predicting the time until his/her next ride based on completed ride experience and alternative modes. A hybrid structure learning method is proposed to firstly design an expert BN based on prior casual dependency relationships between key variables and further improve the dependency structure by implementing data-driven learning algorithms. An empirical case study is conducted on the historical ride data of Kussbus, an on-demand shuttle bus service for cross-border workers in Luxembourg and its Belgium border area. The BN model based on the proposed hybrid structure learning algorithm outperforms the other benchmark models and is similar to MNL. The advantage of the BN model is to uncover more complicated relationships between the factors that are missing in other models. The result suggests that providing subsidies and reducing the total access and egress distance may increase the probability of Kussbus user’s future rides within the next seven days.