Part B: Choose one of Questions 10 or 11

5. We now examine the differences between LDA and QDA.

    (a) If the Bayes decision boundary is linear, do we expect LDA
    or QDA to perform better on the training set? On the test set?

        The QDA has more flexibility, so it will match the training
        set more closely than the LDA. The LDA will perform better
        on the test set because the real relationship is linear, so
        the QDA would have additional bias.


    (b) If the Bayes decision boundary is non-linear, do we expect
    LDA or QDA to perform better on the training set? On the test
    set?

        QDA will still perform better on the training set, but now
        should also perform better than LDA on the test set, since
        QDA will account for the additional degree of freedom in the
        real relationship.


    (c) In general, as the sample size n increases, do we expect the
    test prediction accuracy of QDA relative to LDA to improve,
    decline, or be unchanged? Why?

        Definitely increase. The LDA has an advantage when the
        training set is small because it is less sensitive to the
        fluctuations of those few data. As the size of the training
        set grows, the QDA is able to optimize its coefficients
        well, and assuming the real relationship is not linear, the
        QDA should eventually out-perform the LDA.


    (d) True or False: Even if the Bayes decision boundary for a
    given problem is linear, we will probably achieve a superior
    test error rate using QDA rather than LDA because QDA is
    flexible enough to model a linear decision boundary. Justify
    your answer.

        False. The QDA will likely be biased because it will fit to
        training data that don't completely represent the
        relationship that will be observed in test data.


8. Suppose that we take a data set, divide it into equally-sized
   training and test sets, and then try out two different
   classification procedures. First we use logistic regression and
   get an error rate of 20 % on the training data and 30 % on the
   test data. Next we use 1-nearest neighbors (i.e. K = 1) and get
   an average error rate (averaged over both test and training data
   sets) of 18 %. Based on these results, which method should we
   prefer to use for classification of new observations? Why?

        Definitely 1-nearest neighbor. The logistic regression
        performed more poorly on the training data, to which it has
        been optimized as much as possible, and yet the nearest
        neighbor model performs better over the entire dataset.
        Considering the logistic regression performed even worse on
        the test data, the average error rate of the logistic
        regression over the training and test data is 25%. This
        suggests that the relationship may not even be linear, and
        the nearest-neighbor is a very solid method for modeling
        non-linear classifications, so if the real relationship is
        not linear, it easily explains why the nearest-neighbor
        method would do so much better. Everything here seems to
        point at using the nearest-neighbor.


9. This problem has to do with odds.

    (a) On average, what fraction of people with an odds of 0.37 of
    defaulting on their credit card payment will in fact default?

    (b) Suppose that an individual has a 16 % chance of defaulting
    on her credit card payment. What are the odds that she will de-
    fault?



11. In this problem, you will develop a model to predict whether a
    given car gets high or low gas mileage based on the Auto data
    set.

    (a) Create a binary variable, mpg01 , that contains a 1 if mpg
    contains a value above its median, and a 0 if mpg contains a
    value below its median. You can compute the median using the
    median() function. Note you may find it helpful to use the
    data.frame() function to create a single data set containing
    both mpg01 and the other Auto variables.

    (b) Explore the data graphically in order to investigate the
    associ- ation between mpg01 and the other features. Which of the
    other features seem most likely to be useful in predicting mpg01
    ? Scat- terplots and boxplots may be useful tools to answer this
    ques- tion. Describe your findings.

    (c) Split the data into a training set and a test set.

    (d) Perform LDA on the training data in order to predict mpg01
    using the variables that seemed most associated with mpg01 in
    (b). What is the test error of the model obtained?

    (e) Perform QDA on the training data in order to predict mpg01
    using the variables that seemed most associated with mpg01 in
    (b). What is the test error of the model obtained?

    (f) Perform logistic regression on the training data in order to
    pre- dict mpg01 using the variables that seemed most associated
    with mpg01 in (b). What is the test error of the model obtained?

    (g) Perform KNN on the training data, with several values of K,
    in order to predict mpg01 . Use only the variables that seemed
    most associated with mpg01 in (b). What test errors do you
    obtain? Which value of K seems to perform the best on this data
    set?