Predict Car Prices

The model is trained on data from Ebay car sales data and can predict the car price with an approximation as low as 1814$. This value was obtained by applying a multivariate K nearest neighbors regression with 5 features and a k value of 16.

Best features for univariate K nearest neighbors regression

By sampling a few k values on all the numeric features found in the dataset, the 5 that gave the lowest RMSE were:

• highway_mpg
• curb_weigh
• eng_size
• city_mpg
• width

Optimal settings for multivariate K nearest neighbors regression (number of features and k value)

Based on the first graph different combination of the best features were tested for a range of parameters. The couple of setting that gave the lowest RMSE was with 5 features and a k value of 16.