SMART: A Statistical, Machine Learning Framework for Parametric Risk Transfer
SMART is a project funded by World Bank's Challenge Fund. An initiative of the Global Facility for Disaster Reduction and Recovery (GFDRR) and the UK’s Department for International Development (DFID) that has the purpose to bring innovation to developing countries afflicted by natural disasters like flood, hurricane and earthquake. In this context, SMART tries to address these issues through the application of appropriate machine learning and statistical concepts to develop a new framework for parametric trigger modelling, using as a case study the Dominican Republic. The project covers Thematic Area 2 of the Terms of Reference, entitled “Machine Learning and Big Data for Disaster Risk Financing. Started in 2019 and reaching its conclusion in June 2021, the project focus on two perils: flood and drought.
As part of the deliverables of the project, we were required to provide all source code of the models developed and a web app that would showcase and divulge the main findings of the project.
This repository, is divided in two main folder: MachineLearning, containing the scripts used to create, train and evaluate the machine learning models, and App containing all the files and code used to run the web application.
Inside the machine learning folder, are stored reproducible scripts for the two line research the project focused on:
- Identification of extreme events: Floods and Droughts
- Prediction of milk production
The web App is designed with the intent of getting the final users familiarise with machine learning models and parametric insurance. As of the 3rd of June 2021, the app is still under development and can be found at:
https://luigic.shinyapps.io/SMART/
The final report of the project can be found at:
https://www.gfdrr.org/en/challengefund
The project has also provided the opportunity to produce scientific manuscript published on peer-reviewed journals.
Published Work
Cesarini, L., Figueiredo, R., Monteleone, B., and Martina, M. L. V.: The potential of machine learning for weather index insurance, Nat. Hazards Earth Syst. Sci., 21, 2379–2405, https://doi.org/10.5194/nhess-21-2379-2021, 2021.