Unsupervised embedding and similarity detection of microregions using public transport schedules
This work addresses the need for better spatial data representation in urban planning, specifically for public transport analysis, but it is incremental as it applies existing representation learning to a new data type.
The authors tackled the problem of representing spatial data for machine learning by developing an unsupervised method to embed public transport timetables into vector space, using data from 48 European cities to identify micro-regions with similar transport characteristics and evaluate transport quality.
The role of spatial data in tackling city-related tasks has been growing in recent years. To use them in machine learning models, it is often necessary to transform them into a vector representation, which has led to the development in the field of spatial data representation learning. There is also a growing variety of spatial data types for which representation learning methods are proposed. Public transport timetables have so far not been used in the task of learning representations of regions in a city. In this work, a method is developed to embed public transport availability information into vector space. To conduct experiments on its application, public transport timetables were collected from 48 European cities. Using the H3 spatial indexing method, they were divided into micro-regions. A method was also proposed to identify regions with similar characteristics of public transport offers. On its basis, a multi-level typology of public transport offers in the regions was defined. This thesis shows that the proposed representation method makes it possible to identify micro-regions with similar public transport characteristics between the cities, and can be used to evaluate the quality of public transport available in a city.