Revealing Digitally Invisible Groups through a Machine Learning Approach Using Multi-Source Data

Paper by Wenlan Zhang et al: “Big data has emerged as a critical instrument for urban planning, but is limited by reliability and representativeness. Its availability varies across space, time, and sociodemographics, especially in the Global South. This results in the emergence of digitally invisible groups-individuals who are unable to contribute to and benefit from data-informed decisions-thus exacerbating existing inequalities and further marginalising populations already at risk of exclusion. This study presents an example application, i.e., land use classification in a developing country, combining traditional geospatial data (satellite imagery, night lights, building footprints) with big data (geotagged Twitter, street view imagery) and compares random forest classification results across different data solutions using multiple evaluation methods. Results show residents of informal settlements in Nairobi are underrepresented in geotagged Twitter data, and inaccessible neighbourhoods lack streetview imagery. Relying on a single data source doesn’t improve urban mapping; biases can make some groups digitally invisible. Integrating multiple data sources is likely to mitigate these gaps by capturing different people and places, but this should be guided by a systematic, class-wise comparison rather than indiscriminate aggregation. We recommend primary data collection and participatory mapping to address blind spots and ensure these communities are represented in data-driven decision-making…(More)”.