Rule-Based Support Vector Machine Classifiers Applied to Tornado Prediction

Theodore B. Trafalis1, Budi Santosa1, and Michael B. Richman2

1School of Industrial Engineering, The University of Oklahoma, 202 W. Boyd, CEC 124, Norman, OK 73019
ttrafalis@ou.edu
bsant@ou.edu

2School of Meteorology, The University of Oklahoma, 100 E. Boyd, SEC 1310, Norman, OK 73019
mrichman@ou.edu

Abstract. A rule-based Support Vector Machine (SVM) classifier is applied to tornado prediction. Twenty rules based on the National Severe Storms Laboratory’s mesoscale detection algorithm are used along with SVM to develop a hybrid forecast system for the discrimination of tornadic from non-tornadic events. The use of the Weather Surveillance Radar 1998 Doppler data, with continuous data streaming in every six minutes, presents a source for a dynamic data driven application system. Scientific inquiries based on these data are useful for dynamic data driven application systems (DDDAS). Sensitivity analysis is performed by changing the threshold values of the rules. Numerical results show that the optimal hybrid model outperforms the direct application of SVM by 12.7 percent.

LNCS 3038, pp. 678-684.

Last modified: