Data Mining: Practical Machine Learning Tools and Techniques, Second Edition

The procedure is best illustrated by an example. We will use the familiar

weather data again, but without the playattribute. To track progress the 14

instances are labeled a, b, c,...,n(as in Table 4.6), and for interest we include

the class yesor noin the label—although it should be emphasized that for this

artificial dataset there is little reason to suppose that the two classes of instance

should fall into separate categories. Figure 6.17 shows the situation at salient

points throughout the clustering procedure.

At the beginning, when new instances are absorbed into the structure, they

each form their own subcluster under the overall top-level cluster. Each new

instance is processed by tentatively placing it into each of the existing leaves and

evaluating the category utility of the resulting set of the top-level node’s chil-

dren to see whether the leaf is a good “host” for the new instance. For each of

256 CHAPTER 6| IMPLEMENTATIONS: REAL MACHINE LEARNING SCHEMES

Figure 6.17Clustering the weather data.

a:no a:no b:no c:yes d:yes e:yes a:no b:no c:yes d:yes

e:yes f:no

a:no b:no c:yes d:yes

e:yes f:no g:yes

b:no c:yes

a:no d:yes h:no e:yes f:no g:yes

a:no d:yes h:no c:yes l:yes

b:no k:yes

g:yes f:no j:yes m:yes n:no

e:yes i:yes