We now review k-fold cross-validation.
(A)) Explain how k-fold cross-validation is implemented.
K-fold cross-validation is impemented as follows:
(B) What are the advantages and disadvantages of k-fold crossvalidation relative to:
(i) The validation set approach?
(ii) LOOCV? | Advantages: First, this approach has less bias than the validation approach. Second, this approach yields` different MSE when applied repeatedly due to randomness in the splitting process, while performing LOOCV multiple times will always yield the same results, because we split based on 1 obs.
In Chapter 4, we used logistic regression to predict the probability of default using income and balance on the Default data set. We will now estimate the test error of this logistic regression model using the validation set approach. Do not forget to set a random seed before beginning your analysis.
(A) Fit a logistic regression model that uses income and balance to predict default.
#Load the "Default" data set and produce a summary
data("Default")
summary(Default)
## default student balance income
## No :9667 No :7056 Min. : 0.0 Min. : 772
## Yes: 333 Yes:2944 1st Qu.: 481.7 1st Qu.:21340
## Median : 823.6 Median :34553
## Mean : 835.4 Mean :33517
## 3rd Qu.:1166.3 3rd Qu.:43808
## Max. :2654.3 Max. :73554
#Fit logistic regression model
lrm=glm(default~income+balance, data= Default, family = "binomial")
summary(lrm)
##
## Call:
## glm(formula = default ~ income + balance, family = "binomial",
## data = Default)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.4725 -0.1444 -0.0574 -0.0211 3.7245
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.154e+01 4.348e-01 -26.545 < 2e-16 ***
## income 2.081e-05 4.985e-06 4.174 2.99e-05 ***
## balance 5.647e-03 2.274e-04 24.836 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 2920.6 on 9999 degrees of freedom
## Residual deviance: 1579.0 on 9997 degrees of freedom
## AIC: 1585
##
## Number of Fisher Scoring iterations: 8
(B) Using the validation set approach, estimate the test error of this model. In order to do this, you must perform the following steps:
set.seed(1)
lrm_split = initial_split(Default, strata = "default", prop = .8)
lrm_train = training(lrm_split)
lrm_test= testing (lrm_split)
lrm2 = glm(default~income+balance, data= lrm_train, family = "binomial")
summary(lrm2)
##
## Call:
## glm(formula = default ~ income + balance, family = "binomial",
## data = lrm_train)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.4768 -0.1413 -0.0572 -0.0211 3.7332
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.165e+01 4.905e-01 -23.750 < 2e-16 ***
## income 2.320e-05 5.648e-06 4.107 4.01e-05 ***
## balance 5.647e-03 2.547e-04 22.173 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 2306.8 on 7999 degrees of freedom
## Residual deviance: 1236.3 on 7997 degrees of freedom
## AIC: 1242.3
##
## Number of Fisher Scoring iterations: 8
lrm_probs=predict(lrm2,lrm_test, type="response")
lrm_pred=ifelse(lrm_probs>.5,"Yes", "No")
(lrm_cm=table(lrm_pred, lrm_test$default))
##
## lrm_pred No Yes
## No 1919 55
## Yes 10 16
(1-sum(diag(lrm_cm))/sum(lrm_cm))
## [1] 0.0325
(C) Repeat the process in (b) three times, using three different splits of the observations into a training set and a validation set. Comment on the results obtained.
First run with a 80/30 split.
set.seed(1)
lrm_split = initial_split(Default, strata = "default", prop = .7)
lrm_train = training(lrm_split)
lrm_test= testing (lrm_split)
lrm2 = glm(default~income+balance, data= lrm_train, family = "binomial")
summary(lrm2)
##
## Call:
## glm(formula = default ~ income + balance, family = "binomial",
## data = lrm_train)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.1040 -0.1382 -0.0549 -0.0199 3.7311
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.190e+01 5.313e-01 -22.388 < 2e-16 ***
## income 2.529e-05 6.013e-06 4.206 2.6e-05 ***
## balance 5.776e-03 2.747e-04 21.029 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 2084.0 on 6999 degrees of freedom
## Residual deviance: 1088.3 on 6997 degrees of freedom
## AIC: 1094.3
##
## Number of Fisher Scoring iterations: 8
lrm_probs=predict(lrm2,lrm_test, type="response")
lrm_pred=ifelse(lrm_probs>.5,"Yes", "No")
(lrm_cm=table(lrm_pred, lrm_test$default))
##
## lrm_pred No Yes
## No 2895 71
## Yes 11 23
(1-sum(diag(lrm_cm))/sum(lrm_cm))
## [1] 0.02733333
The error rate using a 70/30 split is 2.73%.
Second run with a 60/40 split
# First run with 70/30 split
set.seed(1)
lrm_split = initial_split(Default, strata = "default", prop = .6)
lrm_train = training(lrm_split)
lrm_test= testing (lrm_split)
lrm2 = glm(default~income+balance, data= lrm_train, family = "binomial")
summary(lrm2)
##
## Call:
## glm(formula = default ~ income + balance, family = "binomial",
## data = lrm_train)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.0727 -0.1394 -0.0557 -0.0206 3.7090
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.186e+01 5.628e-01 -21.073 < 2e-16 ***
## income 2.704e-05 6.402e-06 4.224 2.4e-05 ***
## balance 5.725e-03 2.911e-04 19.668 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 1813.94 on 5999 degrees of freedom
## Residual deviance: 943.73 on 5997 degrees of freedom
## AIC: 949.73
##
## Number of Fisher Scoring iterations: 8
lrm_probs=predict(lrm2,lrm_test, type="response")
lrm_pred=ifelse(lrm_probs>.5,"Yes", "No")
(lrm_cm=table(lrm_pred, lrm_test$default))
##
## lrm_pred No Yes
## No 3863 91
## Yes 13 33
(1-sum(diag(lrm_cm))/sum(lrm_cm))
## [1] 0.026
The error rate using a 60/40 split is 2.6%.
Third run with a 90/10 split.
# First run with 70/30 split
set.seed(1)
lrm_split = initial_split(Default, strata = "default", prop = .75)
lrm_train = training(lrm_split)
lrm_test= testing (lrm_split)
lrm2 = glm(default~income+balance, data= lrm_train, family = "binomial")
summary(lrm2)
##
## Call:
## glm(formula = default ~ income + balance, family = "binomial",
## data = lrm_train)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.5127 -0.1391 -0.0556 -0.0203 3.7293
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.176e+01 5.094e-01 -23.09 < 2e-16 ***
## income 2.315e-05 5.831e-06 3.97 7.18e-05 ***
## balance 5.735e-03 2.648e-04 21.65 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 2198.9 on 7499 degrees of freedom
## Residual deviance: 1156.6 on 7497 degrees of freedom
## AIC: 1162.6
##
## Number of Fisher Scoring iterations: 8
lrm_probs=predict(lrm2,lrm_test, type="response")
lrm_pred=ifelse(lrm_probs>.5,"Yes", "No")
(lrm_cm=table(lrm_pred, lrm_test$default))
##
## lrm_pred No Yes
## No 2408 63
## Yes 10 19
(1-sum(diag(lrm_cm))/sum(lrm_cm))
## [1] 0.0292
The error rate using a 90/10 split is 2.92%
The results identify that the validation estimate of the test error rate can be variable. This variability is a result of random selection of observations for the training set and the validation set.
(D) Now consider a logistic regression model that predicts the probability of “default” using “income”, “balance”, and a dummy variable for “student”. Estimate the test error for this model using the validation set approach. Comment on whether or not including a dummy variable for “student” leads to a reduction in the test error rate.
set.seed(1)
lrm_split = initial_split(Default, strata = "default", prop = .8)
lrm_train = training(lrm_split)
lrm_test= testing (lrm_split)
lrm2 = glm(default~income+balance+student, data= lrm_train, family = "binomial")
summary(lrm2)
##
## Call:
## glm(formula = default ~ income + balance + student, family = "binomial",
## data = lrm_train)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.4737 -0.1402 -0.0558 -0.0206 3.7343
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.106e+01 5.583e-01 -19.810 <2e-16 ***
## income 7.672e-06 9.331e-06 0.822 0.411
## balance 5.724e-03 2.595e-04 22.056 <2e-16 ***
## studentYes -5.672e-01 2.705e-01 -2.097 0.036 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 2306.8 on 7999 degrees of freedom
## Residual deviance: 1232.0 on 7996 degrees of freedom
## AIC: 1240
##
## Number of Fisher Scoring iterations: 8
lrm_probs=predict(lrm2,lrm_test, type="response")
lrm_pred=ifelse(lrm_probs>.5,"Yes", "No")
(lrm_cm=table(lrm_pred, lrm_test$default))
##
## lrm_pred No Yes
## No 1919 57
## Yes 10 14
(1-sum(diag(lrm_cm))/sum(lrm_cm))
## [1] 0.0335
The error rate is 3.35%. The addition of the dummy variable student did not have a significant affect on the estamated error rate.
We continue to consider the use of a logistic regression model to predict the probability of “default” using “income” and “balance” on the “Default” data set. In particular, we will now computes estimates for the standard errors of the “income” and “balance” logistic regression coefficients in two different ways : (1) using the bootstrap, and (2) using the standard formula for computing the standard errors in the glm() function. Do not forget to set a random seed before beginning your analysis.
(A) Using the summary() and glm() functions, determine the estimated standard errors for the coefficients associated with income and balance in a multiple logistic regression model that uses both predictors.
lrm3=glm(default~income+balance, data= Default, family = "binomial")
summary(lrm3)
##
## Call:
## glm(formula = default ~ income + balance, family = "binomial",
## data = Default)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.4725 -0.1444 -0.0574 -0.0211 3.7245
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.154e+01 4.348e-01 -26.545 < 2e-16 ***
## income 2.081e-05 4.985e-06 4.174 2.99e-05 ***
## balance 5.647e-03 2.274e-04 24.836 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 2920.6 on 9999 degrees of freedom
## Residual deviance: 1579.0 on 9997 degrees of freedom
## AIC: 1585
##
## Number of Fisher Scoring iterations: 8
The income standard error is 4.985e-06. The balance standard error is 2.274e-04.
(B) Write a function, boot.fn(), that takes as input the Default data set as well as an index of the observations, and that outputs the coefficient estimates for income and balance in the multiple logistic regression model.
boot.fn = function(data, index){
boot_model = glm(default ~ income + balance, data = data, family = "binomial", subset = index)
return (coef(boot_model))
}
(C) Use the boot() function together with your boot.fn() function to estimate the standard errors of the logistic regression coefficients for income and balance.
boot(Default, boot.fn, 500)
##
## ORDINARY NONPARAMETRIC BOOTSTRAP
##
##
## Call:
## boot(data = Default, statistic = boot.fn, R = 500)
##
##
## Bootstrap Statistics :
## original bias std. error
## t1* -1.154047e+01 -2.143565e-02 4.227110e-01
## t2* 2.080898e-05 -2.321050e-07 5.175976e-06
## t3* 5.647103e-03 1.626487e-05 2.188357e-04
(D) Comment on the estimated standard errors obtained using the glm() function and using your bootstrap function.
The estimated standard errors obtained by the two methods are very close.
We will now consider the Boston housing data set, from the MASS library.
(A) Based on this data set, provide an estimate for the population mean of medv. Call this estimate ˆµ.
data(Boston)
(mean_estimate = mean (Boston$medv))
## [1] 22.53281
(B) Provide an estimate of the standard error of ˆµ. Interpret this result.
(se_estimate = sd(Boston$medv)/sqrt(nrow(Boston)))
## [1] 0.4088611
(C) Now estimate the standard error of ˆµ using the bootstrap. How does this compare to your answer from (b)?
boot.fn2 = function(data, index){
se_estimate = sd(Boston$medv[index])/sqrt(nrow(Boston))
return (se_estimate)
}
boot(Boston, boot.fn2,1000)
##
## ORDINARY NONPARAMETRIC BOOTSTRAP
##
##
## Call:
## boot(data = Boston, statistic = boot.fn2, R = 1000)
##
##
## Bootstrap Statistics :
## original bias std. error
## t1* 0.4088611 -0.0003119504 0.01606233
The standard errors of mean for a single mean and using the bootstrap method are the same.
(D) Based on your bootstrap estimate from (c), provide a 95 % confidence interval for the mean of medv. Compare it to the results obtained using t.test(Boston$medv).
(CI = c(22.53281 - 2 * 0.01671901, 22.53281 + 2 * 0.01671901))
## [1] 22.49937 22.56625
t.test(Boston$medv)
##
## One Sample t-test
##
## data: Boston$medv
## t = 55.111, df = 505, p-value < 2.2e-16
## alternative hypothesis: true mean is not equal to 0
## 95 percent confidence interval:
## 21.72953 23.33608
## sample estimates:
## mean of x
## 22.53281
In the t-test the confidence interval is wider than using boot stap. Using the bootstrap is more precise because it delivers the same confidence level in a narrower range.
Note: I can trust that the average population of medv is between the range of 0.3754231 and 0.4422991. If i repeat this same process with different samples of the same size, I will be wrong only 5%.
Note:The standard error is the standard deviation of the average distribution
(E) Based on this data set, provide an estimate, ˆµmed, for the median value of medv in the population.
(med = median(Boston$medv))
## [1] 21.2
(F) We now would like to estimate the standard error of ˆµmed. Unfortunately, there is no simple formula for computing the standard error of the median. Instead, estimate the standard error of the median using the bootstrap. Comment on your findings.
boot.fn3 = function(data, index){
med_boot = median(Boston$medv[index])
return (med_boot)
}
boot(Boston, boot.fn3,1000)
##
## ORDINARY NONPARAMETRIC BOOTSTRAP
##
##
## Call:
## boot(data = Boston, statistic = boot.fn3, R = 1000)
##
##
## Bootstrap Statistics :
## original bias std. error
## t1* 21.2 0.02055 0.3779922
The average variation of the median of medv is 0.3804337
(G) Based on this data set, provide an estimate for the tenth percentile of medv in Boston suburbs. Call this quantity ˆµ0.1. (You can use the quantile() function.)
(tenth_quan = quantile(Boston$medv, .10))
## 10%
## 12.75
(H) Use the bootstrap to estimate the standard error of ˆµ0.1. Comment on your findings.
boot.fn4 = function(data, index){
tenth_quan_boot = quantile(Boston$medv[index], .10)
return (tenth_quan_boot)
}
boot(Boston, boot.fn4,1000)
##
## ORDINARY NONPARAMETRIC BOOTSTRAP
##
##
## Call:
## boot(data = Boston, statistic = boot.fn4, R = 1000)
##
##
## Bootstrap Statistics :
## original bias std. error
## t1* 12.75 -0.0213 0.5272611
The standard error for the 10th percentile of medv is 0.4950111