P-004 · 2026 · Team project · Industry partner
Lüneburg Public Bus Routing with HVV Timetable Data
A team project with KVG in Leuphana’s Machine Learning Lab. My work focused on GTFS timetable processing, passenger assignment and comparative evaluation of existing services and optimised bus routes.
The planning problem
Bus routes for school journeys must meet arrival windows, capacity limits and constraints on journey duration and transfers. The project compared existing scheduled services with dedicated school-bus routing approaches, including simulated annealing, linear programming and genetic algorithms.
Building the timetable baseline
I developed a pipeline to filter GTFS public-transport timetables for relevant morning services. I checked timetable entries against official schedules, identified missing bus lines and incorporated supplementary records into a consolidated schedule. This provided the reference used to assess alternative routing scenarios.
Assigning passengers to services
I implemented and refined assignment logic that searches relevant stops and checks departure times, transfers and journey duration. I improved transfer indexing and added configurable preferences for fewer transfers, shorter journeys or later departures. I also built compliance reports to flag violations and diagnose why a feasible assignment could not be found.
Comparing routing approaches
I developed the evaluation workflow for comparing the timetable baseline with the team’s routing algorithms. It brings assignment coverage, journey-time compliance and resource-use measures into a shared comparison. I integrated algorithm outputs, resolved identifier mismatches and improved diagnostics for unassigned passengers. I also corrected time-conversion and metric-definition issues so that comparisons used consistent measures.
My contribution and project scope
My contribution centred on data engineering, assignment logic and evaluation. I also prepared baseline methodology and comparison material for the team presentation. The optimisation algorithms were part of the wider team effort. This overview excludes passenger records, precise locations, detailed schedules and numerical results from the private repository.