This project was built to compete for the Vitech Insurance quoting challenge at YHack 2017.
What it does
This program predicts and recommends the appropriate life insurance plan according to the person's profile and needs. While recommending the plan, the program also predicts the premium for each insurance category according to their risk factors and preconditions.
How we built it
The app is built for the Android platform and makes use of Google Maps API for users to look and visualize the general plan demographic which helps the end user in selecting the right plan. The user can make make their personal profile and is registered on the app using firebase authentication service. To recommend the right plan we trained a random forest classifier and four random forest regressors to accurately predict the right plan class appropriate for them, and at the same time also assess the monthly premium.
Challenges we ran into
Training the model in a scalable manner. Handling 1.4 million rows from 3 tables is not an easy task. Creating map visualizations for nearby areas in the region. Handling Python 2 to 3 migrations and deprecated libraries. Designing a smooth UI and UX.
Accomplishments that we're proud of
Successfully importing all the data from the API and training 5 models on it. Create a full user journey from authentication to plan recommendations within 36 hours. Predicting the correct plan with 86% accuracy.
What we learned
Sleep is a myth. Data overlays on map are not straightforward. Predicting continuous data is not as straight forward as categorical data.