Database Reference
In-Depth Information
•
Also, 12 non-correlated sets of experiments, which do not contain any
change, were implemented in order to estimate the actual 1st type error
rate (
) — the “detection” of a change that doesn't occur. This error is
also called “false positive rate” or “false alarm rate”.
α
The expected outcomes of the hypothesis testing on the various types
of changes are described in the following tables.
It is expected that every uncorrelated change implemented will lead to
the associated outcome of the relevant test as indicated in Tables 3 and 4.
The second part of the experiments on artificially generated datasets
evaluated the change detection procedure on time series data. During the
third and sixth periods (out of seven consecutive periods) two changes (in
classification “rules”) were introduced into the database and implemented
disregarding the previous and the consecutive periods. It is expected that
Table 2. Distribution of the artificially generated changes in experiment part1.
Change in
A
Change in
Y
Change in the “Rules”
Total
Database #1
4
2
6
12
Database #2
4
2
6
12
Sum
8
4
12
24
Table 3. Change Detection in Rules (CD).
Change in
A
Change in
T
Change in the “patterns”
(rules)
CD (5%)
TRUE/FALSE
(N/A)
TRUE/FALSE
(N/A)
TRUE
Table 4. Pearson's estimator for comparing distributions of variables (XP).
Database section tested
Change in
A
Change in
T
Change in the
“patterns” (rules)
Candidate
Variables XP(5%)
TRUE
TRUE/FALSE
(N/A)
TRUE/FALSE
(N/A)
Target
Variables XP(5%)
TRUE/FALSE
(N/A)
TRUE
TRUE/FALSE
(N/A)
Candidate & Target
Variables XP(5%)
TRUE
TRUE
TRUE/FALSE
(N/A)
