CRANFIELD UNIVERSITY Eleni Anthippi Chatzimichali ...
CRANFIELD UNIVERSITY Eleni Anthippi Chatzimichali ...
CRANFIELD UNIVERSITY Eleni Anthippi Chatzimichali ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
2.3.2.3 Comparison of validation techniques for the optimisation of the (RBF)<br />
SVM hyperparameters<br />
As thoroughly explained thus far, the prediction power and accuracy of a classifier is<br />
chiefly dependent on the optimisation (tuning) of its hyperparameters. The grid and<br />
surface plots of Figure 2-8 and Figure 2-9 aim at assessing the outcome of the<br />
optimisation process and defining the best technique. Simulations were initiated using<br />
a coarse grid-search; the grid resolution was gradually refined to a grid-step of<br />
, forming a final grid space of 6561 points as presented in Figure 2-8. For<br />
each combination of hyperparameters in the grid, bootstrapping, 10-fold crossvalidation<br />
and LOOCV were applied on the training set.<br />
According to Section 1.6.3, even though LOOCV is a nearly unbiased technique, it<br />
often presents cases of unacceptable high variance, especially when applied to<br />
relatively small datasets such as the ones of case study 1. In addition, LOOCV is a<br />
computationally expensive validation technique that may lead to long execution times<br />
and computationally prohibitive solutions for relatively large datasets (Boardman and<br />
Trappenberg, 2006). On the contrary, 10-fold cross-validation, proved to be the fastest<br />
technique among the three. A great advantage of 10-fold cross-validation is that all<br />
instances within a dataset are eventually used for both training and testing. However,<br />
since the outcome highly depends on the random split into folds, this approach may<br />
lead to training instabilities and relatively high variance (see Section 1.6.2). In<br />
addition, cross-validation proved to be extremely prone to overfitting, especially<br />
during the fine-tuning process. Therefore, this technique proved to be unreliable.<br />
Ensembles of SVMs were also optimised using bootstrapping. Fine grid-search<br />
proved to be extremely fruitful since all overall prediction accuracies increased by a<br />
minimum of 2%. As presented in Section 1.6.4, the probability for any given instance<br />
not being selected in a bootstrap set is approximately 36.8%; thus, bootstrapping<br />
minimises the chances of overfitting (Kohavi, 1995). Indeed, overfitting was no<br />
exception in this case either, however, it was only present in a miniscule number of<br />
instances. Even though bootstrapping is a fairly straightforward method, it constitutes<br />
a computationally demanding statistical procedure that leads to extremely long runs.<br />
The execution times may increase exponentially for larger datasets.<br />
53