i am thrilled to share that a paper co-authored with a pair of my students has been accepted to be read at this year's annual conference of the Royal Geographical Society :-)
our paper is titled Exploring the utility of deep learning in Geographic Information Systems (GIS) to infer socio-economic development, and represents my team's first foray in the intersection of Data Science, AI and GIS. it is scheduled to be read on the 31st of August, in the session on Data and development.
this year's conference will take place as a hybrid event at Newcastle, from the 30th of August till the 2nd of September.
the abstract of our paper reads:
Substantive and reliable data on the economic well-being of a population is crucial to the effective formulation of research and policy. With this information, governments are able to understand and monitor the living standards of their people, evaluate the effectiveness of targeted policies, and make decisions about the allocation of scarce resources.
However, there currently exists a large gap in data available on these measures of economic development, particularly in developing countries, severely hampering future policy efforts.
Due to this, the regions of low socio-economic background are not addressed and the lives of those residing there are negatively impacted.
Hence, this project aims to explore the feasibility of applying remote sensing-based methods of poverty prediction in other countries to the Indonesian context. Specifically, this paper describes an independent research project conducted by a pair of Sixth Form students under the mentorship of a Research Scientist at the National Institute of Education in Singapore. In this project, we aimed to develop a Deep Learning model that is able to output relative wealth index classifications based on the Daytime Satellite imagery of a specific region.
The program utilized capable and established Artificial Intelligence (AI) such as Convolutional Neural Network (CNN). This project involved data preparation, training, fine-tuning our model, extracting our CNN features, and finally conducting ridge regression. We utilised proxy training tasks. With our results, we sought to demonstrate the applicability of this poverty prediction Deep Learning framework to a new geographical context.
The latter involved model training, fine-tuning the model, and finding appropriate datasets with which the model could be trained.
Using the results of this experiment, we hope that the necessary policies can be made to improve socio-economic backgrounds, the likes of which could not be done before.