<p>Bike-sharing services are popular for short-distance trips, and bikes can be rented for a short period and left at convenient locations for the next user. However, the rapid development of the bike-sharing system has caused problems such as a shortage of available bikes during peak hours. Due to the diverse travel patterns of bike-sharing users, it is essential to analyze spatiotemporal travel behavior and accurately predict their travel destinations. This paper proposes a data-informed macro-micro integration approach for bike-sharing destination prediction, which includes a region-level travel demand prediction, individual travel destination prediction, and enhanced strategies with trip chain distributions. A Transformer neural network is built to excavate the user travel preferences and predict the aggregate bike-sharing travel demand. We then design the Stacking ensemble learning approach to predict the users’ travel destinations. The forecast arrival trip demand and knowledge from individual trip chains are adopted to enhance trip destination results. Experiments are conducted with real-world bike-sharing travel records from Beijing, China. The Transformer achieves high accuracy under different spatiotemporal distribution conditions compared to the baseline model, with an average prediction error reduction of 16.44%. With the destination prediction enhanced strategy, the accuracy and F1-score are increased by 34.8% and 27.4%, respectively. Accurately travel destination prediction enables a deeper understanding of users’ mobility patterns and travel demands, thereby facilitating the effective optimization of bike-sharing services.</p>