Biomedical Engineering Reference
In-Depth Information
Learning Curve of Single K−Nearest Neighbor Classifier
0.7
test error
training error
0.65
0.6
0.55
0.5
0.45
0.4
0.35
3
4
5
6
7
8
9
10
k
Fig. 8.
Learning curve of k-nearest neighbor (KNN) classier.
can see from Table 3. Although the combining rules do not make much
dierences from each other, the result from mean-combining rule is shown
to be optimal among the alternatives. Up to now, our classier is based on
only two features: the spectral Mode and Broadness measures. To demon-
strate how an additional measure may aect the classication accuracy, we
add the left Slope into the feature vector and the classication results are
reported in Table 4. It is apparent that both the test and training errors
decrease a lot as the new feature is added (e.g. the test errors drops down
about 6%).
5. Conclusions
The overreaching goal of this detailed analysis was to determine if indi-
viduals with dierent ocular pathologies exhibit quantiable dierences in
their interaction with graphical user interfaces. These distinctions between
Search WWH ::
Custom Search