Turbulence Probability using Principal Component Analysis and Support Vector Machine Approaches

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Wednesday, 20 January 2010: 11:15 AM
B204 (GWCC)
Kimberly L. Elmore, CIMMS/Univ. of Oklahoma and NOAA/NSSL, Norman, OK; and M. Richman

Presentation PDF (66.3 kB)

Support vector machines (SVMs) are a classifier that uses optimal separating hyperplanes (Vapnik, 1996). For the two-class case, optimal separating hyperplanes maximize the distance to the closest point from either class. They provide a unique solution to the separating hyperplane problem and simultaneously maximize the separating margin between the two classes on the training data which, in theory, leads to better classification performance on the test data. The SVM is also useful for determining probabilities of an event, similar to logistic regression.

The data set used here has many candidate predictor variables. Rather than choose a dicrete subset of these predictors, we use a principal component analysis to develop a set of orthogonal synthetic predictors and then choose an SVM model based on bootstrap model selection.