COMPSTAT 2024: Start Registration
View Submission - COMPSTAT2024
A0461
Title: Group anomaly detection for optimizing urban planning of rental bike services Authors:  Lixuan An - Ghent University (Belgium) [presenting]
Bernard De Baets - Ghent University (Belgium)
Stijn Luca - Ghent University (Belgium)
Abstract: In major cities, bike-sharing programs provide a convenient and eco-friendly transportation mode. However, managing and maintaining a large fleet of rental bikes can be logistically challenging and costly. Rental bike rides in Munich over the past five years (2019-2023) from the Munchner Verkehrsgesellschaft (MVG) bike-sharing service are analyzed to optimize the spatial arrangement of rental bike stations and free return regions through urban planning initiatives. Urban planning tasks are solved through the point process model of extreme value theory, a group anomaly detection technique. To identify potential free return regions in non-free return areas, a group anomaly detection task is built based on bike ride end locations. All bike rides ending in a specific bike station region in a non-free return area form a group, where the expected distance from the end location to the nearest station should be close to zero. In this setting, anomalous groups might indicate potential free return regions for urban planning. Furthermore, another group anomaly detection task is aimed at optimizing the distribution of bike stations. In this case, a group refers to all bike rides starting from the same bike station and ending in non-free return areas. Anomalous groups provide valuable insights for improving the distribution of station locations, ensuring better accessibility and convenience for users.