Our goal: "To predict housing rebuild cost from limited attributes with deep learning."
By Levi Stringer, Colin Cumming, Jacob Laframboise, and Nic Merz.
At Queen's Machine Intelligence Neuroevolution Design (QMIND)
In collaboration with Cooperators Insurance Group.
Home owners often want to purchase short temp insurance form their insurance provider, so they can rent out their home on services like airbnb. for this to happen, they must fill out a multiple hour long survey with detailed information about their home and/or have an appraiser come to appraise the value fo the home. having a professional appraiser appraise your house can be a costly process, while potentially taking a substantial amount of time to setup. these costs and delays are good for neither the insuring party or the customer.
A model which requires input of an address and minimal supplemental information and can output an estimation of the rebuild cost of the house. our solution will:
- Use google cloud api's to get the geo coordinates from the address.
- Use google cloud api's to get a satalite and a street view image for the house.
- Use a google cloud api to get all of the points of interest nearby the house.
- Use an API from Open Data Kingston which contains all the neighbourhood boundaries to get the name of the neighbourhood.
- Use a deep learning model to crop the images to focus on the houses.
- Combine the images, and put the images and metadata through a multilayer perceptron and a convolutional neural network.
- Output a prediction of value from the saved training weights of the model.
This work could be deployed as a mobile app that can be integrated with the insurance provider's mobile app. The user would enter the address they want appraised and input some metadata, and then they would get an estimate for the rebuild cost of their house. This would save time and money for both the insurance company and for the customer.