Image Processing Reference
In-Depth Information
FIGURE 6.8: Example of a thermogram of a breast cancer patient (malignant)
#fuzzypartitions classificationrate[%]
2
80.82
3
82.19
4
84.25
5
84.93
6
84.93
7
89.04
8
88.36
9
90.41
10
91.78
11
92.47
12
92.47
13
97.26
14
94.52
15
97.95
TABLE 6.4
Results of breast cancer thermogram classification on training data.
While results on training data provides us with some basic indication of the classification
performance only validation on unseen test data will provide real insights into the general-
isation capabilities of a classifier as normally classification on such unseen patterns is lower
than on previously encountered training samples. We therefore perform standard 10-fold
cross-validation on the dataset where the patterns are split into 10 disjoint sets and the
classification performance of one such set based on training the classifier with the remain-
ing 90% of samples evaluated in turn for all 10 combinations. We restrict our attention on
classifiers with partition sizes of 10 or more as only those achieved good enough classification
performance on the training data.
#fuzzypartitions CRfrom(Schaeferetal.,2007) CRhybridfuzzyapproach[%]
10
78.05
80.27
11
76.57
79.18
12
77.33
80.89
13
78.05
77.74
14
79.53
79.25
15
77.43
78.90
TABLE 6.5
Results of breast cancer thermogram classification on test data based on 10-fold cross
validation.
Search WWH ::
Custom Search