Explainable prediction on migration impacts and modeling trends
MoRE2020 Fellow Juhee Bae, outgoing mobility from University of Skövde to University of Louvain, Belgium
Spatial-temporal predictive analysis have been used in meteorology, environmental science, and in agriculture with the help of statistical analysis and data mining. Recently, neural network was used for human pose modeling and forecasting but needs enough input data. In fact, previous researches note that using traditional time-series forecasting methods is often difficult and challenging because of the various factors that are involved in. Moreover, the model’s reasoning and results are not transparent that makes it hard to understand the process. This is a disadvantage of machine learning techniques.
In this project, we aim to understand the determinants and patterns of population dynamics at a highly detailed spatial scale by developing an interpretable predictive analysis tool revealing the ‘black-box’ on human population dynamics. Our end-user, World Bank, has high interest in not only the determinants and patterns but also the long-run root drivers of international migration. An interactive tool that reveals the influencing factors can support users to make informed decisions about health policy, environmental interventions (e.g., natural disasters), and security. The high-resolution data set we plan to use are free and publicly available with the help of high-resolution sensors (e.g., earth population density, geo-referenced data on conflicts, and global warming) which was not available until recently. With the collaboration of the three partners, our work is novel in building an interpretable and interactive predictive model and tool with the public data set on human population movements.
Collaborating end-users: The world bank, Development Research Group