Comparing Feature Extraction techniques using SVM for Early Fault Classification in NFV context

Networks are adopting virtualization techniques and thus, become large distributed software-driven systems. Ensuring Quality of Service (QoS) in such complex environments is critical and arduous especially now. We need to detect and correct expeditiously the issues as well as to understand systems behavior i.e. need for Root Cause Analysis. In this paper, we propose a comparative study of two Feature Extraction (FE) approaches for Early Fault Classification combined with two Support Vector Machine (SVM) algorithms while having preliminary symptoms in a Network Function Virtualization (NFV) based environment. We use data generated with a stimulus-based approach in such a context, and we compare two existing FE techniques in combination with SVM. Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are applied to features for early fault classification. LDA in combination with SVM leads to an accuracy of 90%.