Alarm sequences forecasting based on sparse Bayesian in communication networks

Abstract

Learning to predict communication faults from alarm sequences is an important, real-world problem in communication networks. There are various methods from the areas of statistics and data mining for this purpose. In order to improve predictive efficiency, we propose a prediction with Sparse Bayesian Method (PSBM) in this paper. Furthermore, we also provide the mathematical formulation of the approach. Compared with Support Vector Machine (SVM) method, the new predictive algorithm not only has the same performance of prediction, but also has more accuracy with fewer predictive errors. In particular, our experimental results show that PSBM has only 70% number errors of SVM in the same test environment.