Inspiration
We are all from the Cary, Morrisville, and Apex area of the Triangle, but we all grew up in different communities and neighborhoods. Even though we lived relatively close to one another, our experiences were not always the same. When we were younger, we often wondered how communities that are so geographically close can feel so different. This question inspired us to investigate the differences between school communities using the ODIS dataset.
At first, we compared schools using their Composite Scores; however, this raised an issue: the score does not tell the whole story. It combines the five domains of Economic, Education, Health, Housing, and Crime. Two schools can have the same Composite Score while experiencing very different challenges.
Because of this, we expanded our analysis and developed STEER (Statistical Twins for Educational Equity & Resources). Using STEER, school administrators can find potential peer schools with similar community-stress profiles and not just similar Composite Scores. We hope this brings meaningful collaboration between schools and encourages resource sharing to solve real-life problems as a team.
What it does
STEER allows users to search for a school and view its Economic, Education, Health, Housing, Crime, and Composite score, and compare those profiles with other schools. STEER is intended to encourage collaboration between schools, not to determine exactly what a school should do. This allows administrators to find potential peer schools that may be experiencing similar situations with resources to share.
How we built it
We began by cleaning the ODIS v3 dataset. This involved standardizing column names to snake case, removing unnecessary information, converting the five domains into numerical values, creating unique display names so schools with identical names could be identified, and addressing duplicate data.
To find potential peer schools, we used StandardScalar to allow comparison between domains at different schools. We then used a k-nearest neighbor machine-learning model to calculate which schools had the most similar stress profiles based on user selection of domains from the list of 5 domains: Economic, Education, Health, Housing, Crime.
Challenges we ran into
The main issue we ran into was deciding how to define "similar." We learned that schools cannot be grouped by similarity in Composite Score or location, so we compared all five domain scores and allowed users to choose the comparison scope and area of interest. We also had to handle duplicate school names, missing values, and difficulties with the nearest-neighbor machine learning model.
Accomplishments that we're proud of
Our machine learning and data science experience mostly comes from coursework, not prior real-world projects, so we're proud we were able to take a public dataset and turn it into a working tool in the time we had. None of us walked in as experts. A lot of our early progress came from figuring out who was strongest where and letting them lead on that piece, rather than everyone trying to do everything.
We're also proud of how disciplined we stayed with the data. It would have been easy to settle on a story we liked, something like "these communities struggle because of X," and then go looking for numbers to support it. We tried not to do that. We let the ODIS data show us what it actually showed, even when that meant dropping assumptions we started with, like the idea that Composite Score alone was enough to compare schools by. That's what pushed us to move past Composite Score and build STEER's multi-domain comparison instead.
The thing we're most proud of is that we didn't just walk away with decent interpretations of the data. We walked away having learned how to actually make one: questioning our assumptions, staying skeptical of a single score that flattens five different domains into one number, and being upfront about what a k-nearest-neighbor comparison can and can't really tell a school administrator.
And beyond that, we're proud of how far we got in one day, going from a question about the neighborhoods we grew up in to a tool other people could actually use.
What we learned
We learned the data management process, including cleaning, organizing, analyzing, and displaying a large dataset. We also learned the importance of understanding the limitations of datasets regarding the conclusions that can be made from it.
Throughout this project, we improved many skills that we are semi-familiar with. This includes how to use feature scaling and nearest-neighbor machine learning to compare schools of similar data profiles.
What's next for STEER: Statistical Twins for Educational Equity & Resources
In the future, we would like to improve STEER by adding more detailed visualizations, clearer explanations of similarity scores, and stronger map-based search features.
We would also like to add data-completeness indicators, downloadable comparison reports, and opportunities for administrators to identify and connect with potential peer schools.
The next step would be gathering feedback from educators, school administrators, and community organizations to determine whether STEER’s peer matches are useful in practice. Our long-term goal is to help schools move from simply identifying community challenges to finding relevant partners for collaboration and resource sharing.
Log in or sign up for Devpost to join the conversation.