Poverty Mapping in India using Machine Learning and Deep Learning Techniques

Abstract

Poverty remains a persistent global challenge, that affects millions worldwide and hinders sustainable development goals. Poverty related data is traditionally collected by an on-the-ground household survey which is conducted once in a few years. Unfortunately, in India, the reach of this conventional data collection method has many limitations, and it is costly, time-consuming, and laborious. The study will assist in identifying areas of poverty and also the levels of poverty which will help policymakers in creating policies that will improve such areas. The research leverages a rich variety of data sources which includes satellite imagery, geospatial data, socio-economic surveys and Point of Interest (POI) data. To extract meaningful patterns and correlations within this diverse dataset, various machine learning and deep learning algorithms such as Decision Tree Regressor, Random Forest Regressor, Convolutional Neural Networks (CNNs) and Multi Layer Perceptron (MLP) are employed. With the help of Random Forest Regressor, the study was able to estimate the poverty at village/town level with a R2-score of 0.778.