Short-Term Demand Forecasting in Mobility-on-Demand Systems: A Systematic Literature Review and Research Agenda

Short-term rider demand forecasting is a foundational operational capability for Mobility-on-Demand (MoD) systems, enabling proactive vehicle pre-positioning, dynamic pricing, and service-level optimization across ride-hailing, bike-sharing, carsharing, and demand-responsive transit platforms. Despite a rapidly growing body of literature, the field lacks a comprehensive and critically structured synthesis of methodological developments, input feature practices, evaluation standards, and unresolved research gaps. This paper presents a Systematic Literature Review (SLR) conducted in accordance with the PRISMA protocol, encompassing 291 peer-reviewed studies published between 2016 and 2025 across transportation engineering, intelligent transportation systems, and machine learning venues—the most comprehensive corpus assembled for this topic to date. The review identifies a clear five-generation methodological succession—from classical statistical and machine learning models through recurrent and convolutional deep learning architectures to Graph Neural Networks and transformer-based models—with signal decomposition methods and probabilistic architectures emerging as distinct 2023–2025 trends. Most significantly, we identify a seventeen-dimension research gap matrix that involves research gaps such as probabilistic demand forecasting, cross-city transfer, decision-focused predict-and-optimize frameworks, etc. Further, six concrete research directions grounded in these gaps are proposed, each accompanied by specific methodological proposals rather than general aspirational statements. The findings underscore the need for standardized benchmarking protocols, open dataset releases with documented preprocessing, and a fundamental reorientation of model evaluation from statistical accuracy metrics toward composite operational, probabilistic, and equity-aware performance objectives.