Robust Classification Via Support Vector Machines

Classification models are very sensitive to data uncertainty, and finding robust classifiers that less uncertainty has raised great interest in the machine learning literature. This paper aims construct \emph{Support Vector Machine} under feature via two probabilistic arguments. The first classifier, \emph{Single Perturbation}, reduces local effect of with respect one given acts as a test could confirm or refute presence significant for particular feature. second \emph{Extreme Empirical Loss}, reduce aggregate all features, which is possible trade-off between number prediction model violations size these violations. Both methodologies computationally efficient our extensive numerical investigation highlights advantages limitations on synthetic real-life data.