Table IV shows the evaluation result. The first row and
the second row show specificity score and sensitivity score,
respectively. Followers are their standard deviation.
As can be seen in the Table IV, in term of sensitivity,
Random Forest shows an excellent result with almost 96%
UML CDs images were correctly classified. This follows
with J48 and SVM with 92.5% and 92% respectively.
On the other hand, in term of specificity. LR performed
the best with 91.4% of correctly classified non-UML CDs
images. SVM performed the worst specificity with 89%.
The results also show that the standard deviation on the
results are relatively small (0.01- 0.05) that indicate the results
are considered reliable (small variation). In summary,
LR performs the best in term of eliminating non-UML CD
images. Accordingly, LR is considered as the best classification
algorithm for our classifier with mentioned-extraction
features.
The confusion matrix in Table IV illustrates the classification
result generated by applying the LR classifier to our
dataset. From total of 1300 images, 1183 images were classified
correctly. 596 out of 650 UML CD images were correctly
predicted as UML CDs. Also 587 out of 650 non-UML
CD images were correctly recommended as non-UML CDs.
On the other hand, among 117 incorrectly classified images,
there was 54 false positives (predicted as UML CDs,