Linear Regression Project

Vehicle Fuel Efficiency & Carbon Dioxide Emissions

A study in factors affecting fuel efficiency and CO2 emissions as well as buidling machine learning models to predict fuel efficiency

Summary

The goal of this project was to demonstrate mastery of data science and machine learning concepts. This started with selection of a problem statement and dataset. I chose to leverage a linear regression model against a dataset on vehicle fuel efficiency from the Environmental Protection Agency (EPA)

Report Navigation

Problem Statement

Goals & Questions

Question 1

Question 2

Question 3

Considerations & Way Ahead

Problem Statement

EPA and other agencies spend large sums of money and time conducting fuel efficiency tests.

Hypothesis: Vehicle MPG Varies by Make/Model.

Vehicle MPG varies by make/model (alternately MPG varies by engine displacement or transmission type) and a machine learning model can be built to predict this. Replacing costly and time consuming tests with a machine learning model would certainly benefit agencies such as the EPA.

Null Hypothesis: Vehicle MPG Does Not vary by Make/Model.

Vehicle MPG does not vary by make/model (MPG does not vary by engine displacement/transmission type) or a machine learning model cannot replace laboratory testing for fuel efficiency.

Goals

Create a machine learning model capable of predicting fuel efficiency based on certain variables
Make the machine learning model accurate enough that it could potentially replace costly and time consuming laboratory tests for fuel efficiency

Questions

First question: Have fuel efficiency standards had an impact on CO2 emissions and overall fuel efficiency over time
Second question: Which factors, based on available data, most affect fuel efficiency
Third question: Can a machine learning model be built to predict fuel efficiency

First Question: Have fuel efficiency standards had an impact on CO2 emissions and overall fuel efficiency over time

Figure 1

Fuel Efficiency over Time

Second question: Which factors, based on available data, most affect fuel efficiency

BLUF Not suprisingly, data analysis shows tight correlation between engine displacement and CO2 emission. However, there were some surprises hidden in trending data over time

Figure 2

Relationship Between MPG and Displacement, Cylinders, and CO2 Emissions

Relationship BetweenDisplacement and CO2 Emissions

Figure 3

Relationship Between Displacement, MPG, and CO2 Emissions

Hidden impacts: Graphs of the Averages of Major Variables over Time

Figure 3 below shows the averages of four major variables over time. Further research into the drastic changes in averages revealed two hidden impacts: federal law and vehicle sales trends

The Energy Independence and Security Act of 2007: raised fule economy standards of America's cars, light trucks, and SUVs to a combined average of at least 35 MPG by 2020 - a 10 MPG increase over 2007 levels

Figure 3

Averages of Major Variables over Time

Vehicle Sales Trends: based on the dataset used, the number of SUVs in 2009 dwarfed all other vehicle types (see Figure 4 below)

Figure 4

Count of Vehicle Type over Time

Can a Machine Learning Model be Built to Predict Fuel Efficiency

BLUF: Yes, a linear regression model shows great promise in predicting a vehicles fuel economy based on a number of variables:

vehicle type (or class)
cylinders
displacement
fuel type

The below images are from a prototype prediction model in Jupyter Notebook. In the future this will be deployed as a webapp.

Figure 5

Instructions for application use

Figure 6

Prediction results based on user input

Considerations

This data does not capture all variables that could affect fuel efficiency and CO2 emissions
Hybrid and electric vehicles may skew data

The Way Ahead

Build a separate machine learning model specifically for hybrid and electric vehicles
Build a separate machine learning model specifically for predicting CO2 emissions based on available data
Deploy a front-end user model on a webpage for educational purposes
Deploy a machine learning model for making predictions over large sets of new data for EPA, DOE, and other Agencies and NGOs