Geology Reference
In-Depth Information
Fig. 8.15 SVM modeling results in the calibration phase
Fig. 8.16 SVM modeling results in the validation phase
determines the capability of the model in predicting discharge values, was calcu-
lated as 0.55 with a CC value of 0.76 and a bias value of 57.1 % during the training
phase. The equivalent values during the validation phase are given in Table
8.1
,
along with the corresponding values of an ANN model.
The results show that combined application of EVT with machine learning can
provide useful information in
flood modeling or any extreme value cases with high
dependency on multi-dimensional input space. A support vector classi
er is used in
this study to separate out the main mass from extreme values above the threshold.
This method is clearly useful in the cases of limited data sets and where there is
dif
nite model from available data sets to model extreme
values. One distinct advantage of this approach is that the modeler can produce
useful synthetic data to analyze features of extreme values and the model can give
details (range) of possible input vectors making extreme values. Although we have
culty in making a de
