Answers are grounded in this paper and cite their sources.
On-demand minibus services offer an efficient solution for enhancing public transport in rural areas through flexible routes and schedules. Public transport providers (PTPs) aim to maximize passenger service with these minibuses, yet they encounter the challenge of a dynamic booking process where requests arrive continuously over time. PTPs have to decide immediately which requests to accept but some accepted requests may tie up the resources for future requests. An informed decision on the impact of requests could enable the PTP to serve more passengers. Therefore, the problem of predicting the request fit for rural on-demand minibus services emerges. We propose a request fit predictor (RFP) framework that leverages supervised machine learning models to predict request fit from displacement labels derived via problem-specific features. A time-dependent threshold adjusts fit requirements over the booking horizon, balancing demand and uncertainty. The modular design integrates these fit predictions with either an exact solver or a heuristic method, guaranteeing feasible real-time acceptance decisions. Tested on 26 months of real data from a German PTP, we show that the RFP improves outcomes primarily by changing which requests are accepted rather than how many: the number of accepted requests increases only marginally, while served passengers rise by 15.98% compared to the status quo and by 9.32% over a sequential solver benchmark. The RFP further sustains a stable service level across the entire booking horizon, substantially outperforming a first-come-first-served solver, and does so without excluding any bus stop from service, unlike a distance-based policy. Practitioners benefit from this paper by getting evidence that in resource-constrained rural systems, acceptance composition, service-level stability, and equitable coverage, rather than raw acceptance volume, are the primary levers for improving on-demand transport. For researchers, it offers empirical evidence on a real rural case study of where an offline-learned decision rule is competitive under sparse demand and rigid time-window constraints.
No takes yet. Share an insight, caveat, or question.
Pirmin Fontaine (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: