ML-FordUsedCars

Ford Used Cars Data Linear Regression versus Decision Tree Regression

Required Software

Name	Version
MATLAB	9.10
Statistics and Machine Learning Toolbox	12.1

Setup

1. Create folder data
2. Download Ford.csv from: https://www.kaggle.com/adityadesai13/used-car-dataset-ford-and-mercedes?select=ford.csv and save into the folder data
3. Open MatLab and run main.m

Alternative for steps 1 & 2, on Linux and maybe Macs:

1. Setup Kaggle API using the instructions here
2. Run the command:

bash getData.sh

Then run step 3

LaTex

Install from here. Then open in TeXworks and run typeset.

Files

File Name	Description
analyseRegression.m	Holds the function analyseRegression so that both models are analysed the same
Appendix.pdf	Extra Data for the poster
Appendix.tex	Tex file to create the latest version of the appendix
FeatureImportance.m	Get the feature importances
GetData.sh	Bash command to download the dataset
GetSummaryStats.m	Function similar to pandas summary stats
Intermediate Results.xlsx	Spreadsheet of all of the intermediate results
IntermediateResultsTableBuilder.py	To create tex code for the Implementation results
KFoldLR.m	Experiment with KFold for Linear Regression
KFoldRF.m	Experiment with KFold for Random Forest
Kuga Hybrid MPG adjustments.xlsx	A file I used to work out whether the hybrid Kuga MPG amount was reasonable
main.m	Main MatLab program
Makefie	File for bash commands to live in
MyLibrary.bib	My bibliography for use by the tex file
OptimiseRandomForest.m	The script used to get the optimum hyperparameters for Random Forest
OptimizeLinearRegression.m	The script used to get the optimum hyperparameters for Linear Regression
Output of OptimizeLinearRegression.txt	The output of the run of OptimizeLinearRegression.m
Output of OptimizeRandomForest.txt	The output of OptimizeRandomForest.m with all of the auto parameters
Output of OptimizeRandomForest2.txt	The output of OptimizeRandomForest.m with it being forced to use the Bag method
Poster.pdf	The final file for the poster
Poster.pptx	The raw file for the poster
predictCombinedMdl.m	Average out the predictions of a cross-validated model
README.md	This readme file
readme.txt	A reduced version of this file
SummaryStats.xlsx	Used to format the summary statistics output from MatLab prior to use in the Poster
testonly.m	Used to only run the analysis of the models

To work out required files use:

[fileList, packageList] = matlab.codetools.requiredFilesAndProducts("main.m")

From https://uk.mathworks.com/help/matlab/ref/matlab.codetools.requiredfilesandproducts.html

Recreate savedvars file

Save variables:

• centre
• mdlLR
• mdlRF
• scale
• test_data
• y_test

Running the models (main)

Download the data to the folder data, then run main.m to run both models off against each other and generate the majority of the graphs:

Histograms of the data Scatterplots of the data against the price Linear Regression residuals Random Forest residuals

Supplementary code

Feature Importance

This is used to see which features the models use (needs main to have run first).

Linear Regression's coefficients Relative feature importance for Random Forest

KFoldLR/KFoldRF

These were used to experiment with using K-Fold validation on the models (needs main to have run first). These require the function in predictCombinedMdl to run.

OptimiseRandomForest/OptimiseLinearRegression

These were used to find the best hyperparameters for the relevant models (needs main to have run first).

Random Forest number of trees versus RMSE