Anticipating Natural Disaster Impacts using Data

Inspiration

We were interested in the world bank data from the start, and had come up with many ideas to pursue. However we were inspired by the recent Earthquakes in Türkiye, and wanted to focus on natural disasters, as this was a recent event. Along with how climate change has been playing a large role in severity of natural disasters over the past few years, and the expected severity of natural disasters because of this in the future. We started off by trying to find possible problems around natural disasters which could possibly be solved using data, we eventually found that a problem for local communities and humanitarian organizations is that they do not have the capacity for data to propel appropriate action, nor have good continuous data to access response needs for a Natural Disaster. We thought that an attempt to use current data on countries, using factors such as poverty rates, electrical infrastructure, food and water security, etc., could be used to help predict possible casualties, displacements, resulting insecurities, etc. in the future. We looked to researching how organizations might try and predict these impacts, and could not find from our preliminary research anything to suggest that this data was being used.

What it does

What the project is suppose to do, is showcase data and predictive results from data sets that we have gathered from many sources (the World Bank being one of them), and visualize these results in an intuitive manner for end users. However what we have is more a proof of concept, given the scale and generality of Natural Disasters.

How we built it

We built the web-app with a combination of JavaScript frameworks (React, d3.js, charts.js), and the predictive models (using Least Absolute Shrinkage and Selection Operator, aka LASSO) made use of R, and Python. Ideally we would have liked to incorporate next.js, mongoDB, and perhaps additional JavaScript frameworks, however we did not have the time to do so unfortunately.

Challenges we ran into

We ran into challenges finding quality data that we could use for the models that we had in mind, we had a lot of questions that we wanted to answer, however we were not able to find either complete data, enough data, or the right kind of data to do so. Unfortunately, the World Bank did not have as complete data as we had been hoping for, being allergic to any datapoint past 2016. We had multiple models that we decided to discard because they proved to be highly inaccurate (8 billion Mean Standard Error), which was unfortunate, however when we looked at the feature matrix for the model, it set many co-efficients to zero, thus showing us that none of the data we had chosen could not accurately predict what we wanted (in this case displacement from natural disasters)

Accomplishments that we're proud of

We are proud to have used our model to quite accurately predict two things, number of people that would be impacted and number of dwellings that would be damaged from an array of different natural disasters in Afghanistan, we achieved a mean absolute error of ~48 for the first model and ~19 for the second model.

We are also proud of how we have implemented animations using the JavaScript frameworks we were working with, as this proved to be a lot more work that we had anticipated. Despite not yet having data properly displayed, the overall UX/UI we are satisfied with.

What we learned

We learned that expectations don't always match with reality, however we learned that it is essential to communicate as a team and constantly update. Maybe we should have used git, or have some central location to share data, because it turns out that trying to send everything over Discord and Instagram chat is inefficient and non conductive to smooth collaborative work. We also learned that we needed to dial back our goals to reach the 24 hour deadline, sacrificing tools with steep learning curves such as databases and graphQL in favor for simpler methods we were more comfortable with.

What's next for Anticipating Natural Disaster Impacts using Data

Honestly, this would be fun to continue granted that we can find better data that we can use. With one of the data sets that we found, it was a large enough data set with enough quantitative variables that allowed us to create a predictive model more effectively. We would have also liked to include more user interaction, allowing a greater flexibility of use for our website. Usage of graphQL and a suitable database would significantly increase the power of our project, as well as including more active statistical analysis beyond our currently limited scope. Overall, expansion, connectivity, and flexibility are the overarching next steps for this project.