Bicycle-sharing services have emerged as an essential component of sustainable urban transportation, yet ensuring balanced distribution of bicycles across stations remains a persistent challenge. This paper presents an IoT-based bicycle-sharing system, developed and simulated using Node-RED, that enables users to unlock bicycles from designated stations, complete trips, and return them to destination stations where automatic cost calculation is performed based on map-based travel distance. Unlike conventional bike-sharing simulations, the proposed system generates a real-time dataset reflecting the concerned region's dynamics, thereby extending beyond static historical records. To enhance operational efficiency, a machine learning module is integrated into the framework to forecast station-level bicycle demand by analyzing inbound, outbound, and net flow patterns. The predictive insights are combined with a threshold-based control strategy to determine whether bicycles should be added, removed, or maintained at specific stations, thereby mitigating imbalance-related inconveniences. The system architecture demonstrates the integration of IoT, data-driven simulation, and machine learning to create a realistic, adaptive, and intelligent service model for smart city transportation. The project highlights how Node-RED simulation can serve as both a prototyping environment and a real-time data generator, supporting future deployments of scalable and user-centric bicycle-sharing services.