Machine Learning-Based Mapping of Slums and Analysis Of Slums Patterns: A Case Study Of Kampala District, Uganda. (Unpublished undergraduate dissertation) Makerere University; Kampala, Uganda.
Abstract
Urban planning and policy-making face considerable challenges as a result of the increasing
urbanization and rise of informal settlements in many emerging nations. Effective urban
management requires accurate and current knowledge of slum regions' topographies. Therefore,
the goal of this research project was to create a machine learning-based method for mapping slums
and examining their landscape patterns.
Sentinel 2 images, a random forest classifier, and primary geographic data like OpenStreetMap
data were all employed to achieve this. The data-driven approach's overall categorization accuracy
was 86%. The Slum pattern assessment also depicted a linear spatial pattern from the distribution
of the slum patches all over the study area. The results further showed that the combination of
machine learning for texture analysis, and landscape metrics such as patch density, fragmentation,
and connectivity allow for a deeper understanding of the landscape patterns of slums and their
implications for urban planning and development.
All the developed methodologies and findings from this research are applicable to similar urban
contexts, aiding decision-making processes and promoting evidence-based interventions for slum
areas worldwide.