Regions experiencing population decline face relatively low demand for public transportation; however, the inconvenience caused by the absence of basic public transport services is more severe than in urban areas. This is largely due to the spatial se...
Regions experiencing population decline face relatively low demand for public transportation; however, the inconvenience caused by the absence of basic public transport services is more severe than in urban areas. This is largely due to the spatial separation between residential areas and essential facilities, as well as the deterioration of fundamental transportation infrastructure such as pedestrian facilities. Consequently, mobility services in depopulated regions must be operated under conditions that differ substantially from those of urban areas. At the same time, transportation service providers encounter increasing financial pressure as declining demand leads to escalating operating losses. In such contexts, it is essential to avoid competition among transportation modes. Rather than fostering intermodal competition, mobility services should be supplied in a manner that enables users to travel to desired destinations at appropriate times, thereby providing the right service, in the right place, and at the right time.
Against this background, demand-responsive transport (DRT) services have recently been increasingly introduced in depopulated regions. However, many of these implementations have been driven by policy decisions or civil complaints rather than by systematic, data-driven analyses that reflect actual regional demand. As a result, the functional relationship between conventional fixed-route bus services and newly introduced DRT services has not been clearly or consistently established. This study addresses this gap by proposing mobility service types that are appropriate to local conditions in depopulated regions through quantitative, data-driven analysis. To this end, a machine-learning-based bi-level framework is developed to complement the limitations of existing public transport systems and to mitigate the deterioration of basic mobility rights.
The study area is spatially disaggregated into 500 m × 500 m grid cells, within which data on population characteristics, transportation conditions, public transport supply and usage, and distances to major facilities are collected to construct the analytical environment. In the first stage of the proposed bi-level framework, a machine-learning-based binary classification model is employed to identify the most appropriate mobility service type — for each cell fixed-route bus, bus-type DRT, or taxi-type DRT. In the second stage, adjacent cells are integrated into operational service zones using DBSCAN-based clustering. This process transforms cell-level service recommendations into practically operable transportation service areas, thereby enhancing real-world applicability. In addition, SHAP-based analysis and partial dependence plots (PDPs) are applied to interpret the influence of key variables and to strengthen the explanatory power of the proposed model.
The results indicate that maintaining conventional fixed-route bus services is appropriate in urban cores and central areas of towns and townships, whereas bus-type DRT is more suitable for low-density peripheral residential areas. In regions with even lower population density and minimal transportation demand, taxi-type DRT is identified as the most appropriate service option. The proposed bi-level framework substantially reduces areas without public transport service provision, as measured by the Bus Coverage Index (BCI), while significantly improving accessibility and convenience, as reflected by the Transit Accessibility Index (TAI). Furthermore, the Fit Index, representing the alignment between transportation supply and demand, is shown to improve markedly. In addition, a clear reduction in overall operating costs is achieved through the reconfiguration of a bus-centered supply structure, demonstrating both expanded user access opportunities and tangible economic benefits associated with policy implementation.
This study contributes to the literature by redefining transportation service provision in depopulated regions not from the perspective of single-mode efficiency, but as an optimal combination of mobility services tailored to regional characteristics. This concept is operationalized through a quantitative, machine-learning-based framework. Moreover, the proposed framework can be readily extended to regions beyond the case study area, highlighting its practical applicability. By presenting a systematic methodology for designing integrated operations of fixed-route buses and DRT services according to local conditions, this study offers valuable insights for the development of effective mobility service strategies in depopulated regions.