282: MSCS Data Mining Dr. Craig Struble
Fall 2002, Final Exam closed book, closed notes, one US Letter cheat sheet, calculators OK 100 points 30 questions
Name:
1
Multiple Choice [2 pts each] Circle the BEST answer. 1. Suppose you have a data set with 35 instances. Using leave-one-out testing, the size of a test set for assessment is (a) (b) (c) (d) (e) 1 2 10 17 34
2. Which of the neural network learning techniques is used for clustering, as discussed in class? (a) (b) (c) (d) perceptron back propogation recurrent Kohonen
3. Which of the following uses sampling to reduce data size? (a) (b) (c) (d) (e) (f) (g) k-means k-medoids k-nearest neighbor CLARA agglomerative nesting fuzzy clustering none of the above
4. A measure of redundancy in data attributes is (a) (b) (c) (d) (e) Z score MAD statistic Pearson's correlation Bayes' rule none of the above
5. Suppose you are gambling with a not so honest friend by betting on coin flips. You know that your friend uses a loaded coin about 10% of the time, and that when the coin is loaded, heads is up 75% of the time. If the coin shows heads 8 times in a row, would you call your friend a cheat if you needed to be at least 90% sure that a loaded coin was being used? (a) (b) (c) yes no not enough information to answer the question
2
6. Given an attribute with the ordinal values {extra small, small, medium, large, extra large}, what is the standardized value for medium? (a) (b) (c) (d) (e) (f) (g) (h) (i) (j) 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95
7. In building a model tree for numeric prediction, the measure used to select the best attribute for splitting is (a) (b) (c) (d) (e) information gain 2 statistic standard deviation reduction median absolute deviation none of the above
8. Taking a class from Dr. Struble is like (a) (b) (c) (d) (e) (f) a day on the beach eating Kopp's custard every other class I've taken having a chronic nightmare riding a scary rollercoaster other (provide your own description)
9. Linear regression is used for (a) (b) (c) (d) (e) dimension reduction numeric prediction classification feature selection none of the above
3
10. In general, the worst case time complexity of association rule mining on n data instances and m attributes is (a) (b) (c) (d) (e) O(2m ) O(2n ) O(2m n) O(2n m) none of the above
11. The COBWEB clustering algorithm describes cluster members with (a) (b) (c) (d) (e) (f) a representative member a probabilistic description the center of the cluster a mathematical function of attributes a graphical representation none of the above
12. A ten-fold cross validation is used to (a) (b) (c) (d) (e) select the best of ten generated models estimate the accuracy of a final model generated with all data sample data when the data set is too large to construct a model waste CPU cycles none of the above
13. Suppose a frequent 4-itemset F has been identified. The maximum number of association rules that can be generated from F is (a) (b) (c) (d) (e) (f) (g) 1 2 4 8 16 not enough information none of the above
14. Principal component analysis is used for (a) (b) (c) (d) (e) dimension reduction numeric prediction classification feature selection none of the above
4
15. The J48 algorithm is used for (a) (b) (c) (d) (e) (f) (g) association rule mining clustering decision tree construction linear regression neural networks bayesian classification none of the above
16. A silhouette plot displays (a) (b) (c) (d) (e) the number of members in each cluster a measure of dissimilarity of members in each cluster a hierarchical clustering a shadow on the wall none of the above
17. Pruning a decision tree during its construction is called (b) (a) (c) (d) (e) postpruning prepruning feature selection dimension reduction none of the above
18. When feature selection is applied to features before executing the mining algorithm, it is called a (a) (b) (c) (d) (e) preselection approach postselection approach wrapper approach filter approach none of the above
19. Fuzzy analysis differs from k-means and k-medoids clustering in that (a) (b) (c) (d) (e) it is a hierarchical clustering technique it is probability based it allows instances to be members of multiple clusters it samples the data none of the above
5
20. The property of being a correlated itemset is (a) (b) (c) (d) (e) downward closed upward closed minimally closed maximally closed none of the above
21. Suppose you are mining data with integral attributes (i.e., containing only integer values). Which distance measure would be most appropriate to use when comparing similarity? (a) (b) (c) (d) rank correlation Manhattan distance Euclidean distance match/mismatch count
22. The 2 statistic is used to test if attributes are (a) (b) (c) (d) redundant outliers normal dependent
23. Nearest neighbor techniques can be used for (a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k) (l) clustering classification prediction association rule mining a and b a and c a and d b and c b and d c and d more than two of a, b, c, and d none of the above
6
24. The first step(s) in the knowledge discovery process are (a) (b) (c) (d) (e) data selection visualization data transformation data mining data cleaning and integration
25. The measure of interestingness used by association rule mining that is based on how frequently an itemset appears in data is called (a) (b) (c) (d) (e) correlation confidence support importance none of the above
7
Short Answer [10 pts each] 26. (a) Why is the na Bayesian classifier na ive ive? (b) What is a Laplace estimator? (c) Why is a Laplace estimator used in na Bayesian classifiers? ive
8
27. Draw and label a picture of a perceptron; i.e., a single node in a feed forward neural network.
9
28. Compare and contrast the 3 -edit rule using the typical mean and standard deviation estimators vs. the median and MAD scale estimators (i.e., Hampel identifier).
10
29. (a) What is one application of the expectation-maximization algorithm? (b) Briefly describe the expectation-maximization algorithm.
11
30. Suppose you are asked to perform data mining for a web based department store (e.g., Amazon.com). The store maintains a log of URL accesses to the site with an associated session ID (which groups together all URL accesses by a single user that occur within time of access, IP address of accessing machine, etc. Furthermore, a database linking the URL to a number of product attributes on the pages. See the sample tables below. The store would like to know if there are specials they can offer their customers based on access habits of customers during a session. For example, should they offer specials on books (which they make little money on), if people buy clothes? How would you mine the data to find patterns that would be useful for making this decision? Discuss each step in the KDD process: data transformation, data selection, appropriate task, model, model and pattern assessment, and result visualization. Be as specific as you can be, but do not be concerned about details of the software used (i.e., telling me you'd write a Java program, Perl script, or use Weka is uninteresting to me). Session ID URL Date Time Machine IP ...
Table 1: Table of URL accesses URL Book Food Computer Music Electronic Clothes ...
Table 2: Table of URL characteristics
12
Additional space for your answer.
13