Information Technology Reference
In-Depth Information
Fig. 18.10
Non-linear fitting surface for the fuzzy model
Syst
vs.
Age
and
Bmi
The foregoing fuzzy modeling provides us with a basis for various such analyses
in which we consider the interrelationships between the variables. Below we sketch
some typical modelings.
18.4
Discriminant Analysis
In discriminant analysis we aim to classify the objects of our data set into groups
according to such previously performed methods as cluster analysis. Hence, for ex-
ample, if we have previously specified three groups for our objects in a space of n
variables, in discriminant analysis we aim to provide a method for classifying our
objects into these groups correctly. From the neural nets standpoint, clustering of
objects and their classification according to the clustering are examples of unsuper-
vised and supervised learning, respectively [2]. We thus “supervise” our model to
carry out correct classification by using representative training data in this task.
Consider our data set above. If we have created such a discrete variable,
Syst3
,
for the systolic blood pressure which has three values, low pressure (
=
1), average
pressure (
3), we could next construct a model for clas-
sifying our persons into these groups correctly according to the variables
Age
and
Bmi
(Fig. 18.11). If our model is sufficiently good, we can also apply it to the other
data sets collected from the same population. Corresponding methods in statistics
for this task are logistic and multinomial regression analyses and Cox's regression
analysis, for example. In computational intelligence we may apply neural nets or
the foregoing fuzzy approach. We sketch first the traditional discriminant analysis.
=
2)andhighpressure(
=
