UChicago Pathways in Data Science · July 2026
Automobile Efficiency Across Eras started as my team's final project in a three-week summer program. At home, I added newer vehicle data to test whether our models still worked on cars from other decades.
01 · How it started
Our class project used cars from 1970–1982. I wanted to see what happened with later data.
At UChicago
I coordinated our four-person team's data cleaning, model comparison, and presentation. We used k-nearest neighbors and multiple linear regression to predict MPG for nearly 400 cars. Weight was the strongest predictor, and k-nearest neighbors performed better.
Back at home
After the program, I added a vehicle sample from 1999 and 2008 and EPA trends through 2024. I kept the sources separate because their testing methods differ.
02 · What I added
The datasets use different testing rules, so I documented what can and can't be compared.
I used the UCI file for historical models, production-weighted EPA data for long-term trends, and later vehicle files to test predictions across years.
I tested models on later years, held-out model families, and another vehicle era. Cross-era predictions were poor, but the datasets also differ in vehicle mix and testing methods.
The project includes six notebooks, a repeatable data pipeline, a report, source notes, and a website.
One result: EPA real-world fuel economy rose from 13.1 MPG in 1975 to 27.2 MPG in 2024. But my later-year tests showed that models trained on older cars didn't reliably carry over to newer ones.
03 · Follow the work
Open a section of the project website below.
Compare the models and try an MPG prediction.
Trace fuel economy, vehicle mix, horsepower, weight, and technology over time.
See how prediction errors change across vehicle eras.
Review source coverage, testing-rule differences, and the limits of comparison.