top of page

A Synthetic Population Model For Locating Target Customers 

Loop Insurance Media (2).jpg

Location: Loop Insurance

ExternalLinkArrow_edited.png

Role: Data Science Intern

Date: November 2020 - April 2021

A statistical model that uses an Iterative Proportional Fitting (IPF) algorithm to generate a synthetic population of Texas based on various demographics and other factors that influence Loop's rating engine

How could this be powerful?

Loop's rating engine calculates customers' premiums for their insurance policies based on relevant factors of the customers' profile including demographics, driving behavior, and more. With an accurate synthetic population model, the proportion of customers that adhere to a certain profile for a specific geolocation can be identified.

Loop Insurance Media.jpg

But what is a synthetic population?

A synthetic population refers to artificial population data that fits a distribution of people and its attributes for a certain area. For the purposes of Loop, my synthetic population encompassed US Census Bureau demographic data for census tracts in Texas

And how did I create a synthetic population?

I used an algorithm known as Iterative Proportional Fitting (IPF) to determine the proportion of individuals for each combination of attributes in my synthetic population. IPF iteratively adjusts the cells of the matrix representing my synthetic population such that the totals along the row and column marginals are adhered to

Screenshot 2023-08-27 at 10.17.40 PM.png
ExternalLinkArrow_edited.png

Takeaways from my experience at Loop Insurance? 

At Loop Insurance, I gained first-hand experience in working in a high-paced startup environment. When taking on the task of building a synthetic population model, I exposed myself to advanced statistics and mathematical concepts well beyond my high school years. I also worked closely with Loop's Chief Data Scientist, Xunge Jiang, who taught me clever techniques in cleaning and analyzing large datasets using programming libraries like Pandas.   

** Please note all code is proprietary to Loop Insurance and therefore not publicly accessible. **

AEnB2Ur5juO1svp3OGX2fnqA1bsA2TWa04sgeF-7NF4k9fCx4z3cF7Yk6FgXOIE0PpuOy9L-52eXmzNwKtwe9CIvx-

Yuvanshu Agarwal @2023

bottom of page