Searching for Points of Interest (POIs) (e.g. schools and hospitals) manually is time-consuming, especially at national level. Our approach provided a way where humans are assisted by machine learning to narrow down the search space of the POIs from high-resolution satellite imagery. However, schools can look very different in different regions, For instance, schools in rural and urban areas often don’t have the same characteristics. Furthermore, the overhead signature may vary significantly from country to country. To overcome this challenge, the parameterization of the school classifier can be further explored for optimal classification to specific geolocations.