In paulemms/datamining: Data mining

Introduction

This is Lab 13 on Data Mining.

First we load the datamining package and the data for the lab.

devtools::load_all()
load('Credit.rda')

x.tr <- x[i.tr,]
x.te <- x[i.te,]
attr(x.te, "terms") = NULL
nx.tr <- make_numeric_data_frame(x.tr, "Class", warn=FALSE)
nx.te <- make_numeric_data_frame(x.te, "Class", warn=FALSE)

fmla =
  Class ~ Checking.200 + CheckingNone + Months + History + PurposeUsed.car + 
    PurposeOther + PurposeFurniture.equipment + PurposeRadio.TV + 
    PurposeRepairs + PurposeBusiness + 
    StatusMale.single + StatusMale.married + GuarantorCo.applicant + 
    Other.plansStores + Foreign

The data set

1. [he dataset is the same as lab 12, with training data in x.tr and test data in x.te. ‘lwo other matrices, nx.tr and nx.te, have the predictors numerically coded, for use with logistic regression. You want to predict Class. Mlusclassification rates
2. ‘These are the proportions of good and bad loans in the test set: Bad Good 0.316 0.684 What is the misclassification rate of a classifier which always reports “Good”? What is the misclassification rate of a classifier which always reports “Bad”? The minimum of these two is the baseline rate. Any classifier which does worse than the baseline is essentially worthless.
3. As in the last lab, construct a pruned classification tree and a k-nearest-neighbor classiher with k=6. What are their misclassification rates on the test set?
4. Construct a linear classifer on the all-numeric version of the training set. What is its misclas- sification rate on the test set?
5. ‘he variable fmla contains a formula with the most important predictors. Use quadratic expan- sion on this formula to build a quadratic classifier. What is its misclassification rate on the test set? Of the four classifiers, which do better than baseline? Mlusclassification costs
6. ‘The variable costs gives the cost to the bank of different types of misclassification: predicted

truth Bad Good Bad 0 5 Good 1 0 What is the average cost (total cost divided by the test set size) of a classifier which always reports “Good”? What is the average cost of a classifier which always reports “Bad”? The minimum of these two is the baseline cost. 11. ‘To minimize cost, the bank should say “Good” only when the probability of “Good” exceeds 5/6. For each of the four classifiers, compute the average cost of this policy on the test set. Which are below baseline? 12. You can now get checked off. Save all of your results for the homework. Logistic regression ‘lhe logistic function is similar to tree: fit = logistic(,) summary (fit) Deviance and misclassification rate are also the same: deviance (fit,,rate=T) misclass(fit,,rate=T) Confusion matrix A confusion matrix cross-classines the predictions of a model and the true responses. The model says “Yes” when the probability of “Yes” exceeds p (which is 0.5 by default). confusion (,,p) It costs is a corresponding table of costs, this will compute the total cost of the classifier on the data set: sum (confusion (,,p)*costs) (Juadratic expansion ‘lo create a formula with quadratic terms added, use one of expand. quadratic (fit) expand. quadratic () expand. quadratic () (Just like expand.cross from lab 10.) expand. quadratic assumes all predictors are numeric. The resulting formula may be very big and cause R to run slowly.

paulemms/datamining documentation built on March 1, 2023, 4:01 p.m.