OpenStreetMap quality assessment using unsupervised machine learning methods

Abstract

The reliability and quality of volunteered geographic information (VGI) continue to be pressing concerns. Many VGI projects lack standard geospatial data quality assurance procedures, and the reliability of contributors remains in question. Traditional approaches rely on comparing VGI to an “authoritative” or “gold standard” dataset to assess quality. This study investigates VGI quality by analysing the OpenStreetMap (OSM) database in Ottawa-Gatineau, focusing on historical map features and contributor data to gain an understanding of how users are contributing to the database, and their ability to do so accurately. Unsupervised machine learning analyses expose a cluster of experienced contributors classified as “OSM validators/experts”, which are then further used to attribute data quality. They are identified through a combination of strong contribution loadings associated with the use and experience of advanced OSM editors, and weaker loadings associated with feature creation and frequency of contributions leading to further correction. Limitations are discussed with implications for future work.