Improved Support Vector Machine for Cyber Attack Detection

Proposes an improved and scalable Support Vector Machine (SVM) algorithm for cyber attack classification by modifying Gaussian kernel to enlarge spatial resolution around the margin through conformal mapping such that separability between attack classes is increased. Based on Riemannian geometrical structure induced by the kernal function. Results show iSVM gives 100% detection accuracy for Normal and Denial of Service (DOS) classes comparable to false alarm rate, training, and testing times. Uses two approaches: feature reduction technique using Generalized Discriminant Analysis (GDA) and improved support vector machine classifier. Combining two approaches yield better performance through feature reduction and classification.