Feature selection of radar-derived attributes with linear programming support vector machines

Tornado circulation attributes/features derived largely from the National Severe Storms Laboratory Mesocyclone Detection Algorithm have been investigated for their efficacy in distinguishing between mesocyclones that become tornadic from those which do not. Previous research has shown several of the attributes do not provide effective discrimination. Moreover, there are strong associations between individual attributes. Despite these drawbacks, applications of artificial neural networks and support vector machines have been successful in discriminating tornadic from pre-tornadic circulations. One of the largest challenges in this regard is to maintain a high probability of detection while simultaneously minimizing the false alarm rate. In this research, we apply a linear programming support vector machine formulation based on the L1 norm to do feature selection on radar-derived tornado attributes (features). Our approach will be evaluated based on probability of detection, false alarm rate, bias and Heidke skill.