Overview

Over 15% of your insurance premium covers fraudulent insurance claims. Assure AI seeks to protect you and insurance companies from this fraud by giving an initial estimation for homes damaged by natural disasters.

With Assure AI you can determine loss in property seconds after a natural disaster, eliminating the need for surveys which can take days and allowing quick response from concerned entities.

Assure AI uses deep learning, real-time property valuations, and mathematical insurance models to accurately value property damage using satellite images and geolocation of the area.

Inspiration

Helping people at scale. Spread awareness about the devastating economical effects natural disasters can have on communities. Seeing as machine

What it does

Following a natural disaster, imagery is collected through drone and aerial surveys. Using a convolutional neural network, the photos are processed and damaged buildings/structures identified. Property damage is then classified into 3 categories (0 - no damage, 1 - partial damage and 2 - completely destroyed) and using the Zillow API, a cost estimation for each individual house is calculated. The total value of the houses are added together to give the amount the insurer has to pay and other various metrics including number of destroyed houses, location, total number of houses are outputted along with a map with damaged houses annotated. Individual houses can be looked up to see if they are damaged or not and optimized insurance calculations are made for insurers to develop long term strategic financial plans.

How we built it

We used Tensorflow and Keras to build the Mask R-CNN framework, express and NodeJS for the back end and react for the front end. We also used the Google Maps and Zillow APIs to get information about regional pricing.

Challenges we ran into

We had to spend a good amount of time figuring out how to have our very different technologies communicate with one another.

Accomplishments that we're proud of

Our machine learning models did very well, and yielded very useful information. We were able to manipulate the different API’s we used fairly well. Lastly we are proud to be able to present this information in a very clean and concise manner to the user through our web app.

What we learned

We learned how to work with React; creating modular components that can communicate with one another. Many of us learnt a lot about the capabilities of machine learning. Overall, we worked well as a team and learnt a lot from one another.

Next Steps

  • Damage Classification
  • Subscriptions for satellite and aerial imagery companies (planet.com, arcGIS, KPMG) to get real-time high-definition photography
  • Simulating restoration process of the destroyed area and showing its potential economic benefits in time lapse, using mathematical model similar to ones summarized in https://www.mat.univie.ac.at/~giordap7/EsmMat.pdf to help insurance companies to better manage risks, or plan for potential disasters.
  • Train the Mask R-CNN framework to more accurately classify damage categories
  • Add more classes to the Mask-RCNN framework to detect damage other than residential buildings
  • Procure mentorship from
  • Deploy web app on cloud computing platforms like Microsoft Azure / AWS

Built With

Share this project:

Updates