<p>The most exigent call of the United Nations in its seventeen sustainable goals is for ending poverty ubiquitously by 2030. Many developed and developing countries are exploring the potential of contemporary techniques to determine the poverty index. The literature indicates that machine learning (ML) techniques are applied to remote sensing satellite images to estimate the distribution of wealth score or poverty index. Deep learning methods with transfer learning are efficient and effective methods in comparison to the other traditional ML methods. Mostly, the countries measure poverty using data collected through ground-level surveys. It is a very important problem with a strong social impact and we decided to review and implement alternate better ways for wealth score prediction. We used the Inception-v3 model which extends its training with satellite images to extract the essential features. There are two modes to transfer learning to generate the feature vectors. Firstly the advanced training of the pre-trained model is performed with both daylight and nightlight images and secondly, the pre-trained model is trained using daylight images only. The feature vectors obtained from this deep learning model are used to train the regression models (Ridge, Ransac, Lasso, and KNN) and to predict the wealth score. The models have provided better results in establishing the correlation ((≈ 0.70) between predicted asset value and the asset value obtained from the survey data. The truthful prediction of wealth score can help legitimate bodies to build policies and take action accordingly. It will support accomplishing the goal of ending poverty in every part of the world.</p>