AD699: Data Mining for Business Analytics Spring 2019 Quiz #1

3. A data scientist has just built a multiple linear regression model with an r-squared value of .7915.  This model includes 10 predictors and it was built with 152 rows of data. What is the adjusted r-squared of this model?

4.  Which of the following statements characterizes the difference between explanatory and predictive linear regression models?

a.   When regression is performed for explanatory purposes, a data partition

must be performed beforehand, so that the person who built the model can check the results against a validation set; predictive linear regression, however, is not associated with data partitioning.

b.   Explanatory regression is done so that an equation can be generated with coefficients and an intercept; with predictive regression, however, the intercept and some of the coefficient terms can be viewed as optional.

c.   With explanatory regression, the person constructing the model will not use more than seven inputs a time; with explanatory regression, however, the number of input variables is theoretically limitless.

d.   When regression is conducted for explanatory purposes, the ‘goodness offit’ is one of the most important metrics considered; when regression is conducted for predictive purposes, however, the main consideration is simply the model’s ability to predict outcomes for new records.

5.    You  are  reviewing a linear  regression model that you have built, and you notice a pattern with the residuals.  For the smaller input values, the residuals tend to be very large, but  as  the  input  values  get  bigger  and  bigger,  the  residuals  become  smaller.    What assumption about linear regression appears to have been violated here?

a.   The rule of arbitrary distributions.

b.   Homoskedasticity.

c.   Parsimony.

d.   Error reduction.

6.   The University Grille on Commonwealth Avenue just released the findings from a three year-study of students’ salad orders to determine the popularity of Caesar and Ranch dressing.  In this study, the ordering habits of 3000 students who have ordered salads were analyzed.  185 of these students never ordered any dressing on their salads. 2100 of the students ordered Caesar dressing, but never ordered Ranch.  What is the probability that a randomly-selected student from this survey ordered Ranch?

7.  When comparing user reviews or recommendations, why might correlation distance be a more appropriate metric to use in some cases than Euclidean distance?

a.   Euclidean  distance is unreliable for subjective measurements because if all the data are measured on the same scale (for instance, 1 through 10) then the values must be normalized prior to analysis.

b.   A user might feel that correlation distance is more effective for capturing the meaning among such relationships because the people making the reviews may not have reviewed the same content (for example, user A might have seen and reviewed 5 movies, while user B saw and reviewed 5 completely different movies).

c.   There  is  no  single,  universal  standard  or  benchmark  by  which  different people base their reviews of a product.   The correlation distance between two reviewers captures the direction of their movement as they go from one product to another.

d.   Because Euclidean distances quickly become unreliable when a large data set is  used  to  generate  them,  something significant (like a movie or product database) will require a more powerful metric such as correlation distance instead.

8.   Based on the information shown below, which of the following statements is true about the distance between the reviews made by Andrew and Brian?

a.   The Euclidean distance between Andrew and Brian is less than the hamming distance.

b.   The Euclidean distance between Andrew and Brian is the same as the hamming distance.

c.   The Euclidean distance between Andrew and Brian is greater than the hamming distance.

d.   The Euclidean distance between Andrew and Brian can be determined with this data, but the hamming distance cannot be determined.

9.  You recently learned that someone built a classification tree using R to explore survival among  passengers  on  the  Titanic,  with  two  outcome variables: “Survived” or “Did Not Survive.”  The predictors in this person’s model were:  Gender (Male or Female), Passenger Class (1st, 2nd, 3rd, or Crew), and Age (a numeric variable).  After you built the model, you discovered that the root node was Gender.  What does this mean?

a.   This  means  that  Gender  can  be  considered  a  conclusive  variable  --  it separated  the  passengers  into  “Survived”  or  “Did  Not  Survive”  without requiring the use of additional decision nodes in the classification process.

b.   Among all the possible splits that the model could have started with, it found that  Gender  did  the  best job of creating homogeneous groupings further down the tree.

c.   Gender is the last variable that the model should consider when creating the data  partition  (assuming  the  rows  have  been  randomized  prior  to  the partition, and aren’t already listed in some logical order).

d.   Of all the variables that were used as inputs, Gender is most likely to appear in the final classification nodes.

10.  Which of the following statements is true about trees, splits, and predictors?

a.   When a tree splits based on a particular predictor, it cannot split again on that same predictor (this is why random forests are preferred to single trees).

b.   A tree may split along values for a particular predictor at one point, and then split again on that same predictor at another point further down the tree.

c.   If a tree creates a split at the root node for a particular predictor, it may split again on that predictor, but only once more.

d.   The relationship among splits and predictors is such that the number of

splits will never exceed the square root of the number of predictors used to build the model.

11.  You recently built a classification tree that had 7 terminal nodes.  How many classification rules can be derived from this tree?

12.  A basketball player named Yao Ming is 7 feet, 6 inches tall.  If he and his wife, Ye Li, had a son, what would someone who believes in “regression to the mean” predict about the eventual height of this son?

a.   The son would likely be taller than Yao Ming, especially if Ye Li is taller than the average woman.

b.   The son’s height could be predicted by an exact equation:  Yao Ming’s height, plus Ye Li’s height, divided by 2 (in other words, the mean of their heights).

c.   The son would be taller than the average person, but not as tall as Yao Ming.

d.   The principle of regression to the mean would not apply in this case, since  Yao Ming’s height is so much greater than the mean height for adult males.

13.  You have been instructed to build a data mining model using the variables shown here, with  the  outcome  variable “incidents” and all of the other variables shown (type, year, period,  and service) as predictors.   However, you  are not allowed to perform any data manipulation or data preprocessing -- in other words, you must run a function on these variables  without  changing any  of  them  at all. Which  of  the  following  data  mining algorithms could be used to accomplish this?

a.   Classification Tree.

b.   K-nearest neighbors.

c.   None of the choices shown here.

d.   Multiple Linear Regression.

14.    Using  the  data  shown in the screenshot below, use a Naive Bayesian approach to generate  a  probability  for  whether  someone  will  play  tennis  on  a  day  with  mild temperature and normal humidity.