A Real World Test of My Used Car Price Predictor That Anyone Can Use
Hello everyone, long time no see! A few months ago, you may have come across my blog where I discussed using Linear Regression to build a model to predict the price of a used car. If you missed it, you can find it here.
To catch you all up, I used Linear Regression to build a model that predicted the price of a used car based off its year, mileage, fuel efficiency and if it was a luxury vehicle. It was cool, but having no prior experience with car purchasing or car ownership, I could only test my model against available web listings, and occasionally help my brother price out his 2015 Honda Fit (accurate to $200). I had run a few light tests on the model to determine its accuracy, and left happy that I had created a useful tool for anyone doing some casual car window shopping.
Since then, hundreds of people have been using it and informed me of some of its shortfalls. For example, SUVs and Pickup Trucks don’t work so well with the model. Similarly, high end luxury cars like Porsches and Maseratis don’t work so well either, mostly due to small sample size, or because Maseratis suck and their resale value goes into the toilet after the first 6 months (there are a number of reasons for this, it’s not entirely the car’s fault, mostly the owner).
Unfortunately, in July, while I was driving my brother’s car home from a Korean Fried Chicken run, I was involved in an accident. You can view the video here (moral of the story, get a dash cam). I’ll spare you the long story, a lot of back and forth with insurance, but the damage was sufficient for Geico to determine the car was a total loss. Not because the car was not drive-able, but because the cost to repair the damage would exceed 75% of the car’s worth. Here was the first real world test for my model. Geico had valued my brother’s car at $11,352.00, and my model had predicted a price of $11,878.00! Barring the actual loss of the car, I was thrilled that my model provided some usefulness to my family!
#data-science #linear-regression #used-cars #used-car-sale-market