A Novel Linear Cellular Automata-Based Data Clustering Algorithm - Foundations on Natural and Artificial Computation - page 77

0 450

900

50 500 950

100 550 1000

150 600 1050

200 650 1100

250 700 1150

300

750 1200

350 800

1250

400 850 1300

450 900 1350

Fig. 7. Successive states of the automata lattice for the Iris dataset

Fig. 7 shows the successive states of the automata lattice. As in the synthetic
case each line represents the tape's line in each iteration. As the Iris dataset
contains three classes we have only employed the black, grey and white colors to
represent the Iris setosa, Iris versicolor and Iris virginica data items, respectively.
The final tape's state is achieved in the iteration 1291 although it is not
obtained a perfect clustering in this case. Only the black zone which represents
the Iris setosa shows a compact state. Some individuals corresponding to the
Iris versicolor (grey cluster) are included in the middle of the Iris virginica zone
(white cluster) and vice versa.
This fact can be clearly observed in the matching matrix shown in Table 1. All
the50Irissetosaitemsareclassifiedinthecorrect cluster, 3 Iris versicolor items
are erroneously classified as Iris virginica, and 1 Iris virginica item is erroneously
classified as Iris versicolor.

Table 1. Iris dataset matching matrix obtained with the proposed algorithm

Iris setosa Iris versicolor Iris virginica

Iris setosa 50 0 0

Iris versicolor 0 47 3

Iris virginica 0 1 49

The obtained results are quite comparable to the ones obtained with a con-
ventional k -means algorithm k = 3. The resulting matching matrix is shown in
Table 2.