Introduction

For this week’s take-home lab, you will work on the same data set from Week 4/5 Take-Home Labs. You will solve the very same problem studied in this week’s in-class lab on a much larger and more interesting dataset. The data contained in the file UCI_Credit_Card.csv contains 30,000 consumer records with 24 different variables. You can read a detailed description of the different fields at the following website:

https://archive.ics.uci.edu/ml/datasets/default+of+credit+card+clients

The description from the UCI says marriage should have levels: Marital status (1 = married; 2 = single; 3 = others) However, there are levels (0,1,2,3). You should treat 0 as unknown. the description from the UCI says Education (1 = graduate school; 2 = university; 3 = high school; 4 = others). However, there are levels 1 to 6 for education. Thus here 5 = 6 = unknown. X6-X11: The measurement scale for the repayment status is: -1 = pay duly; 1 = payment delay for one month; 2 = payment delay for two months; . . .; 8 = payment delay for eight months; 9 = payment delay for nine months and above. However, there are many factors that are -2. This is also unknown. So every unknown you should treat them as NA.

Your task is to build the best possible model for predicting whether or not a consumer will default on their credit card payment for the next month (the last column in the dataset).

Assignment

Perform the following tasks:

• Conduct a training/test split of the data, building a 20% held out test dataset

• Fit the best RF model you can (consider feature selection etc.) to the data to predict consumer default.

• Then plot ROC curves for the logistic regression, SVM, KNN, CART, and RF models, and compare their performance.

• Compute the AUC for the logistic regression, SVM, KNN, CART, and RF models, and compare their performance.

• Provide a summary and discussion of your work in written form (.docx or .pdf) that includes the following:

o Q1 Summarize the model/feature selection process you used to fit your RF model

o Q2 Provide a summary of the fitted RF model (i.e. model summary)

o Q3 Provide performance evaluation of the fitted RF model using confusion matrix.

o Q4 How well do you think the fitted RF model to this dataset works?

o Q5 Using ROC curves and AUC, which one of logistic regression, SVM, KNN, CART, and RF models works better with the dataset over all?

ucicreditcard