Credit based network management by discriminate analysis

Abstract

Nowadays network management becomes a more and more challenging issue and is no longer limited to the routine maintenance of software and hardware. This paper proposes a primitive credit scoring system for the network management. The database contains the records of the abnormal events on a campus network for the past two years, which has been sorted by [1]. Records are analyzed and divided into two classes according to their different probabilities of abnormal behavior by applying Principle Component Analysis (PCA) to obtain the weight of each attribute. Five-fold method is employed to train the Support Vector Machine (SVM) classifier. The result indicates that SVM classifier is effective and the system performs well. The credit scoring system can then give various starting credit scores according to the class the user belongs to. We hope to have finer classification of users to enhance the credit scoring system in future work.