Feature selection of radar-derived tornado attributes with support vector machines

Michael B. Richman*, Budi Santosa, Theodore B. Trafalis

*Corresponding author for this work

Research output: Contribution to conferencePaperpeer-review

1 Citation (Scopus)

Abstract

Tornado circulation attributes/variables derived largely from the National Severe Storms Laboratory Mesocyclone Detection Algorithm (MDA) have been investigated for their efficacy in distinguishing between mesocyclones that become tornadic from those which do not. Using a subset of the MDA variables associated with velocity yields 23 potential predictors. Previous research has shown that the discrimination ability of several of the predictors is not good and the predictor pool has strong associations among subsets of these variables. Despite these drawbacks, applications of artificial neural networks (ANN) and support vector machines (SVM) to the MDA have met with success in predicting correctly pre-tornadic circulations. One of the largest challenges in this regard is to maintain a high probability of detection (POD) while simultaneously minimizing the false alarm rate (FAR). Both ANN and SVM are non-linear classifiers and, accordingly, the use of linear statistics to screen the predictor pool a priori may not be logically consistent. In this research, the impact of removing individual predictors is examined on the training and testing errors. Results were encouraging as exclusion of specific variables had a notable impact on the ability to distinguish accurately the tornadic from the non-tornadic circulations when viewed from misclassification rates, POD, FAR, and Heidke skill. A key finding is that inclusion of the current month number (1= January, 2 = February, ..., 12 = December) in addition to a subset of MDA variables used in SVM is the most accurate set of features tested. This methodology of feature selection outperforms SVM based on the MDA alone, achieving a Heidke skill of 0.844 with a POD of 0.835 and a FAR of 0.135 with more parsimonious models.

Original languageEnglish
Pages1201-1206
Number of pages6
Publication statusPublished - 2005
Externally publishedYes
Event85th AMS Annual Meeting, American Meteorological Society - Combined Preprints - San Diego, CA, United States
Duration: 9 Jan 200513 Jan 2005

Conference

Conference85th AMS Annual Meeting, American Meteorological Society - Combined Preprints
Country/TerritoryUnited States
CitySan Diego, CA
Period9/01/0513/01/05

Fingerprint

Dive into the research topics of 'Feature selection of radar-derived tornado attributes with support vector machines'. Together they form a unique fingerprint.

Cite this