Comparative analysis of classification techniques for network fault management

Abstract

Network troubleshooting is a significant process. Many studies were conducted about it. The first step inthe troubleshooting procedures is represented in collecting information. It’s collected in order to identify the problems.Syslog messages which are sent by almost all network devices include a massive amount of data that concern the networkproblems. Based on several studies, it was found that analyzing syslog data (which) can be a guideline for networkproblems and their causes. The detection of network problems can become more efficient if the detected problems havebeen classified based on the network layers. Classifying syslog data requires identifying the syslog messages that describethe network problems for each layer. It also requires taking into account the formats of syslog for vendors’ devices. Thepresent study aimed to propose a method for classifying the syslog messages which identify the network problem.Thisclassification is conducted based on the network layers. This method uses data mining instrument to classify the syslogmessages. The description part of the syslog message was used for carrying out the classification process.The relevantsyslog messages were identified. The features were then selected to train the classifiers. Six classification algorithmswere learned; LibSVM, SMO, KNN, Naïve Bayes, J48, and Random Forest. A real data set was obtained from aneducational network device. This dataset was used for the prediction stage. It was found that that LibSVM outperformsother classifiers in terms of the probability rate of the classified instances where it was in the range of 89.90%–32.80%.Furthermore, the validation results indicate that the probability rate of the correctly classified instances is >70%.