The Future of Urban Movement: Mastering Short-term Rider Demand Forecasting

Mobility-on-Demand systems currently suffer from fragmented data and inconsistent prediction models, which leads to inefficient vehicle placement and increased congestion. This comprehensive review of 291 peer-reviewed studies establishes a definitive roadmap for short-term rider demand forecasting. It identifies how the field moved from basic statistical models to complex Graph Neural Networks and transformer-based architectures between 2016 and 2025. As urban populations grow, the ability to anticipate where a commuter will be five minutes from now is essential for efficient transport. The researchers measured a clear methodological progression, identifying 17 critical gaps in current systems. They found that while models are becoming more complex, they often lack standardised benchmarking and open dataset protocols necessary for real-world deployment.

The Trajectory of Short-term Rider Demand Forecasting

Over the next decade, this research suggests a move toward 'predict-and-optimise' frameworks. This shift could change how cities manage transport resources by focusing on probabilistic outcomes rather than single-point predictions. This may allow operators to manage uncertainty during extreme weather or major public events. Downstream applications include:

• Dynamic pricing and rebalancing for city-wide bike-sharing.
• Proactive pre-positioning for autonomous ride-hailing fleets.
• Equity-aware routing to ensure service for underserved neighbourhoods.

The data suggests that the next five years will see a move away from simple accuracy metrics. Instead, the field will likely prioritise composite objectives that balance speed, operational costs, and fair access for all citizens. This transition aims to turn raw data into a functional tool for more liveable, efficient urban centres.