As longtime residents of Michigan, we wanted to use our time to develop something that has the potential to increase the quality of life of those around us. Luckily, as a public policy grad student, I have been able to combine my expertise with some fellow computer science students to create an application that will greatly help people in need.

The Problem:

The property tax delinquency rate in the city of Detroit is nearly 50%. This leads to many foreclosed homes and wasted potential. A major factor in a Detroit resident's choice to be delinquent is how much they are getting charged in property taxes. Empirical research has shown, however, that a crumbling real estate market and limited resources has lead to the city assessors continuously over-valuing homes by factors as much as 2000%. Worst of all is that those who can least afford to pay taxes are typically overcharged the most.

Although the topic of over-valuation is not new, never before has a resource been created to help those in need of information the most: The residents of Detroit.

The Solution:

Repraise Detroit allows users to input some basic information about their house and get a "second opinion" on the value of their house. Users can then compare this number against the state equalized value (SEV) and decide whether they should pursue the appeal process. Resources for this process are also given in an easy to digest manner.

How we built it:

We combined a variety of data from the City of Detroit's open data portal,, google geolocations, and Zillow to create a formula for valuing houses using least squares regression. Our backend is written in python which parses through all of the relevant parcel and neighborhood data sets, and joins these together with google and zillow api's to produce a detailed spread of data about each house. From here, we pre-processed the data into an algorithm which produces formulas that estimate market values by neighborhood sectors, using Stata and calculating least squares regression. Our php front end allows users to access these formulas by inputting several pieces of information, including their address, which we match to a neighborhood sector and run through our engine to give an estimated market value of their house.

Challenges we ran into:

We needed to refine our data to subsections of Detroit. The wards supplied in the data were not narrow enough to produce efficient algorithms, so we created a mapping of subsections of wards, or neighborhoods, by joining multiple data sets together. Unfortunately, our neighborhood-to-parcel mappings did not offer a simple "neighborhood" field to query on. We solved this problem by implementing a latitude/longitude check by determining if a point was in a polygon (neighborhood) or not.

An additional obstacle we faced was that our initial batch of data did not have sufficient variables to achieve a model with significant results. Because of this, we incorporated data sets from four different sources [,,] including the use of Zillow's API to pull extra data on each house, such as bedroom/bathroom/floor information.

Accomplishments that we're proud of:

This is the first time I've ever written code, so it's been a pretty crazy experience packing the entire process of developing an application into one weekend. As a team, we're proud that we incorporated different majors to create something that will benefit society.

What we learned:

  • I learned a significant deal about code, and we learned about the open datasets and APIs that are available from the State of Michigan. The computer science students were able to learn about statistical analysis including the use of the statistical suite, Stata, and a real world problem with our real world solution.


*Hodge, T. R., McMillen, D. P., Sands, G. and Skidmore, M. (2016), Assessment Inequity in a Declining Housing Market: The Case of Detroit. Real Estate Economics. doi:10.1111/1540-6229.12126

Share this project: