Information Technology Reference
In-Depth Information
In the traditional case two discriminant functions seem to be sufficient for our
data and the first one is related to
Bmi
and the second to
Age
(Table 18.7). This
model can classify 55.1 % of our cases correctly (Table 18.8). Figure 18.11 depicts
the obtained group centroids according to the discriminant functions.
Table 18.7
Structure matrix in statistical discriminant analysis
Function
1 2
Bmi
.904* -.428
Age
.243
.970*
In a fuzzy model we may apply the foregoing regression methods directly, and
thus we still use the variables
Age
and
Bmi
as inputs, but now
Syst3
is our output.
If we generate the 20 initial fuzzy rules in Table 18.9, we obtain a model which
can classify 80.6 % of the cases correctly (Fig. 18.12 and 18.13) when the initial
outputs are rounded into integers, and thus better outcomes are obtained than in the
traditional case. The fuzzy model is also simpler and more conceivable. Naturally
Fig. 18.11
Three systolic blood pressure groups
