Description
Problem 1 In this problem, we will examine the German Credit dataset that can be found on Webcourses
with the homework in the file SouthGermanCredit.asc. All the column names are in German, but you can
find the English translations of the columns at this site. We are interested in the kredit response, which
indicates if an individual has fulfilled their credit contract. Analyze this dataset by following the steps below:
(a) Load the data using read.table(). Rename the columns with their English names. Split the data
into a training and test set.
(b) Perform a logistic regression using the full set of features. Comment on relevant features. Narrow
down your features into the most relevant predictors. What are they? Create a reduced model using
the set of features you have identified.
(c) Plot an ROC curve and calculate the AUC of your curve for the full and reduced model on both the
training and test set (4 ROC curves in all). Comment on the accuracy and overfitting that you observe
for the full and reduced models.
Problem 2 Analyze the dataset in Problem 1 using LQA and QDA. You should report:
• Summary of each model
• The ROC curve and the AUC of each model
• The comparison among LDA, QDA and logistic regression
1 STA 4364
