Carefully explain the differences between the KNN
classifier and KNN regression methods.
The KKN classifier is used when the outcome is qualitative, and
the classifier predicts the outcome to be the class with most
representations among the k nearest neighbors. On the other
hand, KNN regression methods are used when the outcome is
quantitative. The prediction is made based on the average
value of the k nearest neighbors.
This question involves the use of multiple linear regression on the Auto data set.
(a) Produce a scatterplot matrix which includes all of the variables in the data set.
pairs(Auto)
(b) Compute the matrix of correlations between the
variables using the function cor(). You will need to
exclude the name variable, which is
qualitative.
cor(subset(Auto, select = -name))
## mpg cylinders displacement horsepower weight
## mpg 1.0000000 -0.7776175 -0.8051269 -0.7784268 -0.8322442
## cylinders -0.7776175 1.0000000 0.9508233 0.8429834 0.8975273
## displacement -0.8051269 0.9508233 1.0000000 0.8972570 0.9329944
## horsepower -0.7784268 0.8429834 0.8972570 1.0000000 0.8645377
## weight -0.8322442 0.8975273 0.9329944 0.8645377 1.0000000
## acceleration 0.4233285 -0.5046834 -0.5438005 -0.6891955 -0.4168392
## year 0.5805410 -0.3456474 -0.3698552 -0.4163615 -0.3091199
## origin 0.5652088 -0.5689316 -0.6145351 -0.4551715 -0.5850054
## acceleration year origin
## mpg 0.4233285 0.5805410 0.5652088
## cylinders -0.5046834 -0.3456474 -0.5689316
## displacement -0.5438005 -0.3698552 -0.6145351
## horsepower -0.6891955 -0.4163615 -0.4551715
## weight -0.4168392 -0.3091199 -0.5850054
## acceleration 1.0000000 0.2903161 0.2127458
## year 0.2903161 1.0000000 0.1815277
## origin 0.2127458 0.1815277 1.0000000
(c) Use the lm() function to perform a
multiple linear regression with mpg as the response and all
other variables except name as the predictors. Use the
summary() function to print the results. Comment on the
output. For instance:
- Is there a relationship between the predictors and the
response?
- Which predictors appear to have a statistically
significant relationship to the response?
- What does the coefficient for the year
variable suggest?
lm.fit9 <- lm(mpg ~ .-name, Auto)
summary(lm.fit9)
##
## Call:
## lm(formula = mpg ~ . - name, data = Auto)
##
## Residuals:
## Min 1Q Median 3Q Max
## -9.5903 -2.1565 -0.1169 1.8690 13.0604
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -17.218435 4.644294 -3.707 0.00024 ***
## cylinders -0.493376 0.323282 -1.526 0.12780
## displacement 0.019896 0.007515 2.647 0.00844 **
## horsepower -0.016951 0.013787 -1.230 0.21963
## weight -0.006474 0.000652 -9.929 < 2e-16 ***
## acceleration 0.080576 0.098845 0.815 0.41548
## year 0.750773 0.050973 14.729 < 2e-16 ***
## origin 1.426141 0.278136 5.127 4.67e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 3.328 on 384 degrees of freedom
## Multiple R-squared: 0.8215, Adjusted R-squared: 0.8182
## F-statistic: 252.4 on 7 and 384 DF, p-value: < 2.2e-16
Given the significance of the overall model, there appears to be a
relationship between the predictors and the response. The variables that
are statistically significant predictors of mpg are
displacement, weight, year, and
origin. The coefficient of the year variable suggests that
for each unit increase in a car’s model year, its miles per gallon is
estimated to increase by 0.75.
(d) Use the plot() function to produce
diagnostic plots of the linear regression fit. Comment on any problems
you see with the fit. Do the residual plots suggest any unusually large
outliers? Does the leverage plot identify any observations with
unusually high leverage?
par(mfrow = c(2,2))
plot(lm.fit9)
The Residuals vs Fitted plot indicates that the data violates the assumption of homoscedasticity. Additionally, the plot indicates that the data is non-linear given the curved pattern in the residuals. From the Residuals vs Leverage plot, there appears to be no leverage points seeing that no observations cross the long-dashed grey line, which represents Cook’s distance.
(e) Use the star and : symbols to fit linear regression models with interaction effects. Do any interactions appear to be statistically significant?
lm.fit9i1 = lm(mpg ~ horsepower*displacement, Auto)
summary(lm.fit9i1)
##
## Call:
## lm(formula = mpg ~ horsepower * displacement, data = Auto)
##
## Residuals:
## Min 1Q Median 3Q Max
## -10.9391 -2.3373 -0.5816 2.1698 17.5771
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 5.305e+01 1.526e+00 34.77 <2e-16 ***
## horsepower -2.343e-01 1.959e-02 -11.96 <2e-16 ***
## displacement -9.805e-02 6.682e-03 -14.67 <2e-16 ***
## horsepower:displacement 5.828e-04 5.193e-05 11.22 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 3.944 on 388 degrees of freedom
## Multiple R-squared: 0.7466, Adjusted R-squared: 0.7446
## F-statistic: 381 on 3 and 388 DF, p-value: < 2.2e-16
lm.fit9i2 = lm(mpg ~ displacement*weight, Auto)
summary(lm.fit9i2)
##
## Call:
## lm(formula = mpg ~ displacement * weight, data = Auto)
##
## Residuals:
## Min 1Q Median 3Q Max
## -13.8664 -2.4801 -0.3355 1.8071 17.9429
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 5.372e+01 1.940e+00 27.697 < 2e-16 ***
## displacement -7.831e-02 1.131e-02 -6.922 1.85e-11 ***
## weight -8.931e-03 8.474e-04 -10.539 < 2e-16 ***
## displacement:weight 1.744e-05 2.789e-06 6.253 1.06e-09 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 4.097 on 388 degrees of freedom
## Multiple R-squared: 0.7265, Adjusted R-squared: 0.7244
## F-statistic: 343.6 on 3 and 388 DF, p-value: < 2.2e-16
lm.fit9i3 = lm(mpg ~ acceleration*cylinders, Auto)
summary(lm.fit9i3)
##
## Call:
## lm(formula = mpg ~ acceleration * cylinders, data = Auto)
##
## Residuals:
## Min 1Q Median 3Q Max
## -13.2257 -3.1788 -0.7045 2.4031 17.4642
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 31.37192 5.27599 5.946 6.13e-09 ***
## acceleration 0.73498 0.33724 2.179 0.0299 *
## cylinders -1.84692 0.85564 -2.159 0.0315 *
## acceleration:cylinders -0.11179 0.05806 -1.926 0.0549 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 4.895 on 388 degrees of freedom
## Multiple R-squared: 0.6097, Adjusted R-squared: 0.6067
## F-statistic: 202 on 3 and 388 DF, p-value: < 2.2e-16
When assessing the models above, we observe that the interaction
terms between horsepower and displacement and
between displacement and weight are
statistically significant, while the interaction term between
acceleration and cylinders is insignificant.
However, it’s worth noting that the significance levels of these
interactions may change when including additional predictors in the
models.
(f) Try a few different transformations of the variables, such as log(X), √X, X2. Comment on your findings.
reference9 <- lm(mpg ~ acceleration, Auto)
summary(reference9)
##
## Call:
## lm(formula = mpg ~ acceleration, data = Auto)
##
## Residuals:
## Min 1Q Median 3Q Max
## -17.989 -5.616 -1.199 4.801 23.239
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.8332 2.0485 2.359 0.0188 *
## acceleration 1.1976 0.1298 9.228 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 7.08 on 390 degrees of freedom
## Multiple R-squared: 0.1792, Adjusted R-squared: 0.1771
## F-statistic: 85.15 on 1 and 390 DF, p-value: < 2.2e-16
lm.fit9t1 <- lm(mpg ~ poly(acceleration, 5), Auto)
summary(lm.fit9t1)
##
## Call:
## lm(formula = mpg ~ poly(acceleration, 5), data = Auto)
##
## Residuals:
## Min 1Q Median 3Q Max
## -17.2209 -5.2976 -0.9565 4.7597 22.7506
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 23.4459 0.3516 66.689 < 2e-16 ***
## poly(acceleration, 5)1 65.3340 6.9608 9.386 < 2e-16 ***
## poly(acceleration, 5)2 -18.7482 6.9608 -2.693 0.00738 **
## poly(acceleration, 5)3 6.0643 6.9608 0.871 0.38418
## poly(acceleration, 5)4 20.7577 6.9608 2.982 0.00304 **
## poly(acceleration, 5)5 -5.3550 6.9608 -0.769 0.44218
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 6.961 on 386 degrees of freedom
## Multiple R-squared: 0.2148, Adjusted R-squared: 0.2046
## F-statistic: 21.12 on 5 and 386 DF, p-value: < 2.2e-16
lm.fit9t2 <- lm(mpg ~ log(acceleration), Auto)
summary(lm.fit9t2)
##
## Call:
## lm(formula = mpg ~ log(acceleration), data = Auto)
##
## Residuals:
## Min 1Q Median 3Q Max
## -18.0234 -5.6231 -0.9787 4.5943 23.0872
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -27.834 5.373 -5.180 3.56e-07 ***
## log(acceleration) 18.801 1.966 9.565 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 7.033 on 390 degrees of freedom
## Multiple R-squared: 0.19, Adjusted R-squared: 0.1879
## F-statistic: 91.49 on 1 and 390 DF, p-value: < 2.2e-16
lm.fit9t3 <- lm(mpg ~ sqrt(acceleration), Auto)
summary(lm.fit9t3)
##
## Call:
## lm(formula = mpg ~ sqrt(acceleration), data = Auto)
##
## Residuals:
## Min 1Q Median 3Q Max
## -18.034 -5.601 -1.027 4.713 23.184
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -14.177 4.008 -3.537 0.000453 ***
## sqrt(acceleration) 9.582 1.017 9.424 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 7.053 on 390 degrees of freedom
## Multiple R-squared: 0.1855, Adjusted R-squared: 0.1834
## F-statistic: 88.81 on 1 and 390 DF, p-value: < 2.2e-16
When comparing the reference model to the three models testing
various transformation methods on acceleration, we notice
that the transformations improve the multiple R-square in all three
instances. From the polynomial transformation, we also see that the
significance level of accelaration fluctuates as we
increment the power to which it is raised. However, no polynomial terms
beyond the 4th order have significant p-values.
This question should be answered using the Carseats data set.
(a) Fit a multiple regression model to predict
Sales using Price, Urban, and
US.
lm.fit10a <- lm(Sales ~ Price + Urban + US, Carseats)
summary(lm.fit10a)
##
## Call:
## lm(formula = Sales ~ Price + Urban + US, data = Carseats)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.9206 -1.6220 -0.0564 1.5786 7.0581
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 13.043469 0.651012 20.036 < 2e-16 ***
## Price -0.054459 0.005242 -10.389 < 2e-16 ***
## UrbanYes -0.021916 0.271650 -0.081 0.936
## USYes 1.200573 0.259042 4.635 4.86e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.472 on 396 degrees of freedom
## Multiple R-squared: 0.2393, Adjusted R-squared: 0.2335
## F-statistic: 41.52 on 3 and 396 DF, p-value: < 2.2e-16
(b) Provide an interpretation of each coefficient in the
model. Be careful—some of the variables in the model are
qualitative!
The output reveals that Price and US are
significant predictors of Sales, unlike Urban.
More specifically, for every $1000 increase in price, sales at each
location decrease by 54 units. Additionally, locations in the US sell
about 1,201 units more than locations outside the US.
(c) Write out the model in equation form, being careful
to handle the qualitative variables properly.
The estimated regression equation is: \(Sales
= 13.043469 - 0.054459 \cdot Price - 0.021916 \cdot Urban_{Yes} +
1.200573 \cdot US_{Yes}\)
(d) For which of the predictors can you reject the null
hypothesis \(H_0:
β_j=0\)?
As mentioned in (a), Price and US are
significant in the model, why we can reject the null hypothesis for
these two predictors.
(e) On the basis of your response to the previous question, fit a smaller model that only uses the predictors for which there is evidence of association with the outcome.
lm.fit10e <- lm(Sales ~ Price + US, Carseats)
summary(lm.fit10e)
##
## Call:
## lm(formula = Sales ~ Price + US, data = Carseats)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.9269 -1.6286 -0.0574 1.5766 7.0515
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 13.03079 0.63098 20.652 < 2e-16 ***
## Price -0.05448 0.00523 -10.416 < 2e-16 ***
## USYes 1.19964 0.25846 4.641 4.71e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.469 on 397 degrees of freedom
## Multiple R-squared: 0.2393, Adjusted R-squared: 0.2354
## F-statistic: 62.43 on 2 and 397 DF, p-value: < 2.2e-16
(f) How well do the models in (a) and (e) fit the
data?
Both models have a multiple R-squared of 0.2393, meaning that the models
can explain 23.93% of the variance in Sales. This is
generally not a great fit.
(g) Using the model from (e), obtain 95% confidence intervals for the coefficient(s).
confint(lm.fit10e)
## 2.5 % 97.5 %
## (Intercept) 11.79032020 14.27126531
## Price -0.06475984 -0.04419543
## USYes 0.69151957 1.70776632
(h) Is there evidence of outliers or high leverage observations in the model from (e)?
par(mfrow = c(1,1))
plot(hatvalues(lm.fit10e))
From the plot above we see an outlier with an index around 50. Using
the which.max() function below, we find that the exact
index for the outlier is 43.
which.max(hatvalues(lm.fit10e))
## 43
## 43
This problem involves simple linear regression without an intercept.
(a) Recall that the coefficient estimate βˆ for the
linear regression of Y onto X without an intercept is given by (3.38).
Under what circumstance is the coefficient estimate for the regression
of X onto Y the same as the coefficient estimate for the regression of Y
onto X?
The coefficient estimates for the regression of X onto Y and Y onto X
respectively will be the same if the relationship between them is
perfectly linear. In other words, if \(\sum
x^{2} = \sum y^{2}\), the coefficients will be the same.
(b) Generate an example in R with n = 100 observations in which the coefficient estimate for the regression of X onto Y is different from the coefficient estimate for the regression of Y onto X.
set.seed(123)
n <- 100
X <- rnorm(n)
Y <- 2*X + rnorm(n)
lm_XY <- lm(X ~ Y + 0)
coef(lm_XY)
## Y
## 0.3975655
lm_YX <- lm(Y ~ X + 0)
coef(lm_YX)
## X
## 1.936372
sum(X^2)
## [1] 83.30737
sum(Y^2)
## [1] 405.7547
We observe different coefficients as a result of the difference between \(\sum X^{2}\) and \(\sum Y^{2}\)
(c) Generate an example in R with n = 100 observations in which the coefficient estimate for the regression of X onto Y is the same as the coefficient estimate for the regression of Y onto X.
set.seed(123)
n <- 100
X <- rnorm(n)
X_sum_sq <- sum(X^2)
Y_sum_sq <- X_sum_sq / n
Y <- sqrt(rep(Y_sum_sq, n))
lm_XY <- lm(X ~ Y + 0)
coef(lm_XY)
## Y
## 0.09905014
lm_YX <- lm(Y ~ X + 0)
coef(lm_YX)
## X
## 0.09905014
sum(X^2)
## [1] 83.30737
sum(Y^2)
## [1] 83.30737
This time, the coefficients are the same (0.09905014). As explained in (a), this is because \(\sum x^{2}\) is equivalent to \(\sum y^{2}\), which in this case are both 83.30737.