Online Support Vector Machines that are Nonconvex

Abstract

The nonconvex online Support Vector Machine (SVM) method (LASVM-NC) we provide in this research is based on the Ramp Loss and has a good capacity to reduce the impact of outliers. We then suggest an outlier filtering technique (LASVM-I) based on approximating nonconvex behaviour in convex optimization, once more in the context of online learning. These two methods are based on LASVM-G, an unique SVM technique that may generate precise intermediate models through repetitive processes by taking advantage of the duality gap. We offer experimental findings that show the value of our frameworks in attaining appreciable resilience to outliers in noisy data classification where a large number of training examples are incorrectly categorised.