Socioeconomic Classification Using Satellite Imagery and Machine Learning

Abstract

Socioeconomic classification using satellite imagery is an emerging way of monitoring different social classes dwelling in a heterogeneous urban society. As it has proved to be effective in addressing numerous social challenges, developing countries demand such hassle-free mechanisms to keep a close check on the growth rate of these social groups. This research is an attempt to perform this challenging task by using machine learning techniques, with the economic hub of Pakistan, Karachi, as the study area. The study caters the issue of unavailability of very high resolution (VHR) satellite data by using Google basemaps. Furthermore, acquisition of ground truth data was also accomplished through one of the leading and legitimate real estate web portals. The processed VHR raster data of Karachi is classified into three major social classes i.e. lower, middle and upper class, using the grid-based approach with random forest classifier via semi-supervised classification technique. According to the findings, the populated areas of Karachi are constituted by 10%, 11%, and 8% of citizens belonging to the middle, lower, and upper classes, respectively. Classification results demonstrate that a major population of people belonging to the lower class dwell in the west and malir districts, while the upper and middle classes predominate the south and central districts of Karachi. However, all three socioeconomic strata have been shown to make significant contributions to the population of the east district. The results obtained show overall accuracy of 78.2%, which is higher than the accuracy of similar work done in past. The study paves way for automated socioeconomic classification, opening doors for further experimentation in this domain.