$24
1. Consider the following dataset:
married
education
income
creditLine
cardCategory
no
college
low
10k
Blue
yes
college
low
5k
Gold
no
college
low
10k
Blue
yes
highSchool
middle
7k
Silver
yes
graduate
middle
7k
Silver
no
highSchool
high
5k
Red
no
college
middle
10k
Gold
a. Compute the coverage of each item set listed below. (1 pt.)
Item Set
Coverage
education = highSchool, cardCategory = Red
___
married = no, income = low, creditLine = 7k
___
b. Write down every association rule that could be generated from the 2-item set listed below, regardless of whether or not there are actually any instances of that rule in our given dataset. Hint: You should be able to generate 3 rules. (1.5 pts.) married = no, cardCategory = Blue
c. Compute the accuracy of each rule listed below. Express accuracy as a fraction (e.g., 2/3, 2/2, etc.), NOT as a decimal number (e.g., 0.67, 1.0, etc.). (1.5 pts.)
Rule
Accuracy
If married = yes then income = middle
___
If married = no and education = college
then creditLine = 10k and cardCategory = Blue
___
If _ then cardCategory = Red and married = yes
___
1
Name: ___________________________ 26 points possible
2. The dataset shown below is posted on Canvas (along with this assignment) as creditBinary.csv. Run the Prism algorithm on it in Weka specifying cardCategory as the decision attribute. List the classification rules that are produced (you can just include a screenshot of your Weka output). Then work out the Prism algorithm by hand starting with a rule for cardCategory = Blue to show what classification rules you would get; who knows, they might be different than what Weka produces! SHOW
ALL OF YOUR WORK!!! (6.5 pts.)
If there is a tie between 2 attributes, choose the attribute that comes first in the table as listed from left to right (e.g., education comes before creditCardDebt). This will make it easier on the grader (i.e., multiple possible solutions won’t have to be considered!).
married
education
income
creditCardDebt
cardCategory
yes
highSchool
ge50k
low
Blue
yes
highSchool
ge50k
high
Blue
no
highSchool
ge50k
low
Blue
no
college
lt50k
low
Gold
no
college
lt50k
high
Gold
yes
college
lt50k
low
Gold
yes
highSchool
lt50k
high
Gold
no
college
ge50k
high
Gold
no
highSchool
lt50k
low
Gold
yes
college
ge50k
high
Blue
2
Name: ___________________________ 26 points possible
3. Consider the dataset shown below where the decision attribute is paidCash. Assume that attribute weights wmilk, wbeer, wdiapers, and wchips (corresponding to attributes boughtMilk, boughtBeer, boughtDiapers, and boughtChips, respectively) are all initialized to 2. If Ɵ is 2, α is 2, and β is 0.5, what will the attribute weights (i.e., wmilk, wbeer, wdiapers, and wchips) be after one iteration of the Winnow algorithm? YOU MUST SHOW YOUR WORK in computing these values; otherwise, you will receive NO CREDIT! (2 pts.)
boughtMilk
boughtBeer
boughtDiapers
boughtChips
paidCash
x1
0
1
0
1
0
x2
1
1
0
0
1
x3
0
0
0
1
1
x4
0
1
0
0
0
Final values: wmilk = ___ wbeer = ___ wdiapers = ___ wchips = ___
3
Name: ___________________________ 26 points possible
4. Consider the dataset given below where the decision attribute is the one labeled z. Build a kd-tree where k = 2. No partial credit will be given unless you SHOW
YOUR WORK! (8.5 pts.)
When computing medians, if you have a real number, round .1 to .4 down to the next integer, and round .5 to .9 up to the next integer (e.g., round 2.5 to 3, round 2.3 to 2, etc.).
When processing the non-decision attributes, process them in alphabetical order (i.e., x before y).
x
y
z
1
5
green
2
8
blue
2
10
red
3
20
blue
4
20
green
5
30
red
6
40
blue
7
50
green
8
60
red
4
Name: ___________________________ 26 points possible
5. Consider the dataset given below where the decision attribute is the one labeled class. Show how k-means clustering using k = 3 would cluster the instances on attributes a and b assuming that the initial cluster centers you start with are (2, 4), (5, 6), and (8, 1). SHOW ALL OF YOUR WORK!
Use Manhattan distance for your calculations. When computing centers, if you have a real number, round .1 to .4 down to the next integer, and round .5 to .9 up to the next integer (e.g., round 2.5 to 3, round 2.3 to 2, etc.).
Do NOT draw a graph showing the final clusters; simply specify what the clusters will be in terms of what each cluster’s center is and what instances from the dataset will be in each cluster. (5 pts.)
a
b
c
class
2
4
11
true
5
6
5
false
8
1
7
false
7
3
4
true
4
10
8
true
3
0
3
true
9
8
1
false
5