A First Course in FUZZY and NEURAL CONTROL

8.5. INTEGRATED PEST MANAGEMENT SYSTEMS 285

Figure 8.25. Clustering approach

Referring to Figure 8.25, while class objects A and B are clearly separable,
clusters C and D represent groups of class objects that overlap and are therefore
inseparable. Features representing class types A and B, that are distinct, can
easily be trained using the classical feedforward multi-layered neural networks
whereas, objects in clusters C and D can only be determined using fuzzy logic-
based approaches. In a fuzzy approach, the clusters are partitioned optimally,
and a set of fuzzy ìIf...then...î rules are generated. These rules provide a
basis for pattern classification using ANFIS.
Fuzzy subsets, in the form of membership functions used in ANFIS, are
shown in Figure 8.26.

Figure 8.26. Classification scheme

The features listed in Table 8.19 are typical of the input variables needed to
train ANFIS. A total of 222 training patterns from 13 different insect species
were used to train ANFIS. Figures 8.27ó8.36 show the membership functions
before and after training. The range over which each of the input variables is
defined was obtained by looking at the minimum and maximum values of each