Biomedical Engineering Reference
In-Depth Information
Figure 3. Appearance percentage in five GAs
Figure 4. Comparison of performance using all the variables and selected variables
10
1
All variables
Variables found by GAs
10
0
1
10
10
0
50
100
150
200
Epochs
The number of training and test data samples
associated with each rule is also listed in Table
4; this gives us an idea of the importance of each
rule and the distribution of the data.
Regression rules summarised in Table 4 can
also be written in the antecedent/consequent
format. For example, Rule 2, which is executed
most of the time for both the training and test
data, can be rewritten as
However, the above derived
y
' needs to be
mapped back to the normal range using the reverse
function of equation (7) to obtain the GNMM
simulated dispersion coefficient
y
:
(18)
•
IF
t
1
≥ 0.5939
These regression rules could provide environ-
mental managers or field response teams with a
rapid assessment tool for identifying values of
the longitudinal dispersion coefficients required
•
AND
t
2
≥ 2.0104
•
AND
2.1432 ≤
t
3
≤ 4.1393
•
THEN
y
' = −1.7143 + 1.3519
t
3
− 0.3154
t
3
2
Search WWH ::
Custom Search