• No se han encontrado resultados

Ambiente Núcleo

D) Secuencia para la construcción de la visión de gobierno como mito.

a difference of a very small fraction between summation scores of grades, the one with

larger value will be declared and it will not indicated the closeness of the competition. In

case of the majority vote, if different features have different winners then it may result in

a tie as we have seen in case 3 of G-II and case 6 of G-III. A tie reflects that features from

multiple grades behave differently. So majority vote does not always have a winner but

reflects competition among grades for certain cases; in case of a it shows the complexity

involved in classifying a grade. cases.

In the next section, we create grade profiles based on the similarity scores and discuss

which features have been able to perform well in predicting the correct grade.

5.6

Discussion

Figure 5.28 shows a grade profile for two cases of G-I testing data plotting similarity

scores for features as described in previous section. It can be seen that features 1, 2 and

5 provide high similarity scores for both cases with the correct grade where as feature

4 provides the lowest similarity score. For feature 4, G-I similarity was comparatively

higher with the correct grade. Feature 3 is the most inconsistent feature as it was able to

classify case 2 but classified case 1 as G-III. It can also be observed that there is signifi-

cant differences between G-I similarity scores compared to G-II and G-III in general for

all features where G-I was chosen. That is why both maximum sum of similarity and ma-

jority vote performed well for G-I. Another observation is that scores for G-II and G-III

remained very close to each other in more features. We conclude that features 1, 2 and 5

5.6. Discussion 153

(a) Case 1

(b) Case 2

Figure 5.28: Grade profile for two cases of G-I

Figure 5.29 shows grade profiles for six test cases of G-II. Previously we have seen

that both classification methods perform poorly and G-II is not clearly distinguishable

from other grades. However, there are some interesting observations that we can make by

looking at the Figure 5.29. Feature 1 is able to classify the correct grade for cases 2, 3,

4 and 6 and case 5 is narrowly mistaken as G-III. Feature 2 correctly classified the grade

for cases 3 and 6 where as for cases 2, 4 and 5 it was very close to classifying the correct

grade. Feature 5 only classified correctly for case 1. In the majority of the cases where

G-II was not classified correctly, it was classified as G-III. This is generally the case in

real world scenarios, as G-II and G-III are considered very close to each other and chances

of false classification remain high. We conclude that only feature 1 is able to classify the

correct for majority of the G-II cases (4 out of 6) while other features remain inconsistent

5.6.

Discussion

154

(a) Case 1 (b) Case 2

(c) Case 3 (d) Case 4

(e) Case 5 (f) Case 6

5.6. Discussion 155

Figure 5.30 shows grade profile for six test cases of G-III with their similarity scores

for each feature of each grade. It can be seen that feature 3 always classified correctly

except for case 1. Feature 5 also classified correct grade for 4 out of 6 cases (cases 2,3,4,

and 5). In case of features 1 and 2, G-III scores were slightly less than G-II where as in

case of feature 4, G-III was also falsely classified as G-I for cases 2, 5 and 6. We conclude

that features 3 and 5 are best suited for classifying G-III correctly for the majority of the

5.6.

Discussion

156

(a) Case 1 (b) Case 2

(c) Case 3 (d) Case 4

(e) Case 5

5.6. Discussion 157

Table 5.16 shows a summary of the features against correctly classified grades. The

features that were able to classify the grade correctly for test cases have been highlighted.

The results indicate that features performed differently for the three grades. Feature 3

is only significant in case of G-III classification and did not perform well for any other

grade. Similarly, feature 2 only performed well in case of G-I and classified false grade in

all other cases. Our results indicate that various features based on different regions of the

same spectra may provide different information and some may be helpful in classifying a

grade correctly while others may not be useful as explained before. It can also be seen that

zGT-II fuzzy sets based on interval data from spectral regions may be useful in extracting

important information regarding grade classification problems where both inter and intra

variabilities are involved.

Features 1 and 2 have the same peak value but they do not behave identically, although

they have similar results in some cases. This shows that a feature with a common value

can still be useful and may provide useful information for classification.

Table 5.16: Summary of grade profiles

Grades Feature 1 Feature 2 Feature 3 Feature 4 Feature 5

Correctly classified / Total test cases

I 2/2 2/2 1/2 2/2 2/2

II 4/6 2/6 2/6 2/6 2/6

III 0/6 3/6 5/6 1/6 4/6

Table 5.17 shows the summary of the results in terms of correct classification of grades

by both the summation and the majority vote method for all test cases. It can be seen that

summation method has performed well for G-I and G-III test cases. The majority vote

method has performed well for G-I test cases and has shown reasonable results for G-III