A modular machine learning Platform for Resilience Inference Measurement and Enhancement

Keywords: Disaster Resilience, Machine Learning, web platform
Session Type: Virtual Paper Abstract
Day: Saturday
Session Start / End Time: 2/26/2022 09:40 AM (Eastern Time (US & Canada)) - 2/26/2022 11:00 AM (Eastern Time (US & Canada))
Room: Virtual 21

Abstract

Disasters have been affecting human life in myriad ways. Just in 2020, the global pandemic showed the vulnerability that humans face. Even in the midst of such peril, more than 250000 people were affected by 10 types of climatic disasters just in 2020 alone - with the percentage of disasters on the rise. It would be much easier to plan the building up of the community resilience to battle such vulnerability if resilience indexes were readily available to access. Hence, this project aims to use the iRIM model to quantify such resilience and provide a web platform to visualize it. This model uses three dimensions to calculate resilience – exposure, damage, and recovery. The resilience groups would be assigned at first based on these variables using standard deviation. Bayesian network, Principal Component Analysis would be used to consider and validate these previously obtained groups based on socio-economic variables. The SHELDUS (Spatial Hazard Events and Losses Database) dataset from Arizona State University, Census data, NLCD (National Land Cover Database), NED (National Elevation Dataset), and NFHL (National Flood Hazard Layer) are used in this study. The end product will visualize resilience indexes based on counties along with classification accuracy. The tool will also accommodate the user to be able to choose between representing variables in facets where required, providing a lot of flexibility. This tool will enable planners and researchers alike to mark out vulnerable communities at the county level and come up with proper strategies to counter disasters, increase community resilience.