Mining Linguistic Trends from Time Series - Data Mining: Foundations and Practice - page 57

STEP 14: The maximally fuzzy patterns generated from Step 13 are output
as fuzzy linguistic trends that are ( mh ), ( m,h,h,M ), ( h,h,h,H,H,H )and
( M,h,h,h,H,H ).

4 Experimental Results

In this section, the experiments made to show the performance of the pro-
posed method are described. They were implemented in Java at a personal
computer with Intel Pentium IV 3.20 GHz and 512 MB RAM. The dataset
used in the experiments is a set of synthetic control-chart time series from
The UCI KDD Archive [5]. The dataset contains 600 examples of control
charts synthetically generated. The six classes are normal , cyclic , increasing
trend , decreasing trend , upward shift ,and downward shift . Each time se-
ries has sixty data points. One time series of each class was selected to make
the following experiments.
Experiments were first made to show the relationship between numbers of
linguistic trends and minimum support values. The sliding-window size was
set at ten and the number of membership functions for angles is ten, with
five for both positive and negative angles. Results for the class of decreasing
trend are shown in Fig. 2.
From Fig. 2, it is easily seen that the number of linguistic trends decreased
along with the increase of the length of large patterns except for L 2 and L 3 .
Finding L 2 and L 3 was the main effort in the mining process, which was
consistent with previous study [6, 7, 10].

25

20

15

10

5

0

L1 L2 L3 L4 L5 L6 L7 L8 L9 L10

sup = 1% sup = 2% sup = 3% sup = 4% sup = 5%

Fig. 2. The relationship between numbers of linguistic trends and min. sup