Berkeley Haas - Professional Certificate in Machine Learning and Artificial Intelligence - Module 11 - Practical Application 2

• Background
• What drives the price of a car?
• CRISP-DM
  • Overview of Data
  • Data Cleaning and Transformation
  • Fitting Models
  • Coefficient Analysis
• Next Steps

Background

This is a demonstration of application of skills learned during Part 2 of the program; Modules 6 to 11. Specifically, demonstrates execution of one round of CRISP-DM framework.

What drives the price of a car?

The dataset comes from Kaggle. The original dataset contained more data but was trimmed to 426K cars to ensure speed of processing. The goal is to understand what factors make a car more or less expensive. As a result of the analysis, we provide clear recommendations to your client -- a used car dealership -- as to what consumers value in a used car.

CRISP-DM

CRISP-DM framework phases are detailed in following Jupyter Notebook. You will find in there all the details and steps used to complete Business Understanding, Data Understanding, Data Preparation, Modeling and Evaluation.

Technologies and runtime libraries used include: Python, Pandas, Seaborn and scikit-learn.

Overview of Data

• There is a lot of missing values. 72% to 40% for some features.
• Outliers are omnipresent:
  • Car's year model information has bogus entries.
  • Some odometer values are off the charts, like 10 million miles.
  • Price has plenty of 0-values but also irrational values, like $3,736,928,711.
• The target univariate -- car sale price -- is not normally distributed.

Data Cleaning and Transformation

• Removed cars older than year 2000. Also added new column 'age'.
• Removed cars with mileage over 200k miles.
• Keep entries where price is in range: ~500 to ~300,000.
• Convert price to logarithmic representation to get normal distribution.
• Dropped columns with no relevant data, or with majority of missing data.
• Dropped some categorical columns that would just increase number of features. We can come back to them later.
• Left with 130,077 entries down from 426,880.

Fitting Models

Following models were trained:

• Linear Regression.
• Ridge Model with Default Alpha.
• Ridge Model with Optimal Alpha using Grid Search.
• Linear Regression with SFS selection based on Lasso estimator.

All above models performed nearly identically. $R^2$ score around 0.52 and MSE around 0.30.

Coefficient Analysis

It is clear even without the SFS selection that Car's age (year) and odometer reading (mileage) are the key drivers of car's price.

• Linear Regression Coefficients
• Coefficients after SFS

Other features don't quite influence the sales price.

Next Steps

It would be good to explore other categorical features that were initially dropped, especially condition. Encode them and see how they influence sales price.