A dataset of severe weather covariates was applied to both a logistic regression and a support vector machine to determine each model's ability to distinguish between tornadic and non-tornadic severe weather outbreaks. A 21 X 21 grid at 18 km spacing centered on the given outbreak was used from 24 hour forecasts from the WRF model, which was initialized using NCEP/NCAR reanalysis data. Nineteen covariates were selected initially for analysis, and the use of permutation testing over the 441 gridpoints yielded p-values for each gridpoint, which were used to determine which covariates had the largest discrimination ability. Seven covariates resulted from these permutation tests, including the lifted condensation level (LCL), surface based convective inhibition (CIN), 0-1 km bulk shear, the product of surface based convective available potential energy (CAPE) and 0-1 km bulk shear, storm relative environmental helicity at 0-1 km and 0-3 km, and 0-1 km energy helicity index (EHI). 50 tornado outbreaks and 49 non-tornadic severe weather outbreaks were analyzed with these covariates, and a jackknifing procedure was applied to the training and testing of these models to ensure each model type used all data for training and testing. The performance of these models was analyzed through contingency table statistics, with the ratio of the probability of detection (POD) and the false alarm rate (FAR) used to rank the model performance. Models were run using all 7 variables and with different combinations of the 7 to determine the optimal set of covariates. In general, the support vector machine outperformed logistic regression for most model runs.