← back to katsuma.ca ← Projects
A model that predicts what a used car should sell for. I trained a linear regression on more than 100,000 Craigslist listings, then checked it against real cars for sale to see if it held up.
Given the details of a car, it estimates a fair asking price. It landed on an R² of 0.691 and an average error of about $3,781.
A model can look good on the data it was trained on and fall apart in the real world, so I ran it against live listings on Autotrader. On common cars it came within 11 to 13 percent, which is close enough to be useful when you are trying to tell a fair price from a bad one.