Databases Reference
In-Depth Information
milk
bread
Low
Middle
High
Low
Middle
High
1
1
0
356 0 3 6 0
0
6
10
14
18
Quantity
Quantity
cookies
beverage
Low
Middle
High
Low
High
1
1
0
5
8
10
12
15 16
20
0
9 10
15
Quantity
Quantity
Fig. 7.
The final set of membership functions
fitness value) is then output as the membership functions for deriving fuzzy
association rules.After the evolutionary process terminates, the final set of
membership functions for each item is shown in Fig. 7.
After the membership functions are derived, the fuzzy mining method pro-
posed in [5] is then used to mine fuzzy association rules from the quantitative
database.
7 Experimental Results
In this section, experiments made to show the performance of the proposed
approach are described. They were implemented in Java on a personal com-
puter with Intel Pentium 4 2.00 GHz and 256 MB RAM. 64 items and 10,000
transactions were used in the experiments. In each data set, the numbers of
purchased items in transactions were first randomly generated. The purchased
items and their quantities in each transaction were then generated. An item
could not be generated twice in a transaction. The initial population size
P
is set at 50, the crossover rate
p
c
is set at 0.8, and the mutation rate
p
m
is
set at 0.01. The parameter
d
of the crossover operator is set at 0.35 according
to [3] and the minimum support
α
is set at 400.
After 500 generations, the final membership functions are apparently much
better than the original ones. For example, the initial membership functions
of some four items among the 64 items are shown in Fig. 8.
In Fig. 8, the membership functions have the bad types of shapes that are
defined in the previous section. After 500 generations, the final membership
functions for the same four items are shown in Fig. 9. It is easily seen that
the membership functions in Fig. 9 is better than those in Fig. 8. The two bad
