Causal direction inference for network alarm analysis

Automatic alarm analysis is important for network operation. Numerous alarms from different layers of a network may be caused by one single fault. Finding the correct causal direction between two sets of correlated alarms helps to locate the original fault correctly. Causal direction inference can be taken as a task of feature extraction. Generalized Gaussian Distribution (GGD) is used in this work to approximate the distributions of the observations and the unit entropy of GGD is extracted to determine the causal direction. Experiments of the novel method gives satisfactory results on the data from real networks. (C) 2017 Elsevier Ltd. All rights reserved.