Re-engineering

Post-modelling #

Re-engineering #

After running the base models with initially feature engineered dataset, we saw low performing results and then re-engineered.

Contextual Anomaly Detection #

Odometer: ​ #

• Standard car odometer should have max 300,000 miles.​
• Trimming data with 75th quantile + 3*IQR, ~268,000, as a cut reduced skewness from 3.04 to 0.3​. Only new car should have 0 odometer so trimmed otherwise​.

Price:​ #

• Dropped cars below $1000 and greater than $200,000 that​ are in the extreme ranges not fit for our analysis purpose​
• Max car price was three billion dollars ​

MSRP:​ #

• MSRP is the manufacturer’s suggested retail price (list price) ​
• Dropped MSRP < car price as MSRP should be higher than​ used car selling price​

Code Reference

Re-engineering​