Biomedical Engineering Reference
In-Depth Information
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
Control
#1
#2
#3
Fig. 6.
Boxplot for Broadness measures of multifractality.
Neither theoretical guidance or empirical evidence is absent to convince
the choice of linear classier and other parametric classiers for our PRB
measurements. Therefore, the nonparametric model, like k-nearest-neighbor
classier, is preferred in this problem for the sake of better modeling ac-
curacy. As usual, it is very easy to build a nonparametric mode with poor
predictive properties if the model is not tuned very well. To get around
this, cross-validation (CV) strategy is often used to ensure the relative good
model is selected. The idea of CV is to divide the dataset into the training
set and test set. The former is used to estimate the model parameters and
the latter is used to validate the predictive accuracy of the model. In our
problems, the original PRB measurements are divided into two parts for
each group, one of them is assigned to training set and the other is used
to test the trained classier. The training set includes a 90% randomly se-
lected sample of each group from the whole datasets and the rest is taken
to be the test set.
Search WWH ::
Custom Search