From Boardings to Routes: Designing Feeder Bus Networks from Entry-Only Smart Cards

This research develops a demand-driven procedure for designing feeder bus routes using entry‑only smart‑card transactions, GIS, capacitated clustering, and a genetic‑algorithm formulation of the Fixed‑Start Open Multiple Travelling Salesman Problem (GAMTSP). It addresses the operational challenge of converting large volumes of boarding records into feeder bus networks that respect capacity limits and improve service efficiency without costly field surveys. From 1,967,955 boarding records we inferred feeder origins to a suburban train station using a 90‑minute transfer window aligned with operator fare policy and applied an activity filter (≥3 passengers/hour). Candidate stops were aggregated to existing stops and grouped by a demand‑based, capacity‑constrained clustering algorithm (cluster capacity ≤50 passengers/hour) with centroid snapping to operational nodes. Route alternatives (1–5 routes) were generated by a modified GA for the fixed‑start open MTSP; each configuration was repeated for 30 runs to assess solution stability. Initial extraction yielded 321 candidate stops; filtering reduced this to 87 stops (633 passengers/hour), aggregated to 74 stops and finally 13 capacitated clusters. GAMTSP produced feasible route sets: the two‑route alternative achieved the lowest mean tour length (29,256m) versus the single‑route mean (30,739m), a 4.8% reduction; per‑route lengths fell substantially with higher route counts averaging 14,628 for two routes and 10,414 for three, though total vehicle‑kilometres increased. GA repetitions exhibited low variance for the preferred alternatives. By embedding this feeder design approach within an integrated, user-centric, multimodal mobility services, this offers planners a scalable, survey‑less tool to integrate feeder buses into multimodal mobility platforms, balance operator costs with passenger convenience, and provide a data foundation for KPI analysis in urban transit planning.