Models

Models #

• Our goal is to predict the sale price of a used car, which is a supervised regression problem. We pick our models base on two considerations, flexibility (accuracy) and interpretability.​
• We value model accuracy over interpretability because:​
• The industry we are in doesn’t require we provide explanation for the decision we make.​
• Features of our used car dataset are easy to understand, thus making it easy for us to debug the model even without high model interpretability.​

Our Model Selection #

• Linear Regression​

• Support Vector Regression with linear kernel​

• Decision Trees Ensemble method ​

• Bagging Trees (Random Forest)​

• Boosting Trees

Initial Model Selection: ​ #

• Linear Regression is not flexible enough to capture all the variance of the model​
• SVR would be very slow to train. (SVR training time scale badly with large number of training sample)​
• Ensemble Trees would be the best method as it is flexible and has decent interpretability​ ​

Our Approach #

• Train and tune all the models and compare the models’ accuracy​
• Select the model with the best metric scores​

Our Metrics #

• R squared: the proportion of the variance explained by the model​
• Root Mean Squared Error
• Mean Absolute Proportional Error