Feature selection with linear programming support vector machines and applications to tornado prediction

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. 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 is evaluated based on the indices of probability of detection, false alarm rate, bias and Heidke skill. The results are compared to those performance indices obtained after applying branch & bound and sequential forward selection procedures.