Feature selection for classification of BGP anomalies using Bayesian models

Traffic anomalies in communication networks greatly degrade network performance. Early detection of such anomalies alleviates their effect on network performance. A number of approaches that involve traffic modeling, signal processing, and machine learning techniques have been employed to detect network traffic anomalies. In this paper, we develop various Naive Bayes (NB) classifiers for detecting the Internet anomalies using the Routing Information Base (RIB) of the Border Gateway Protocol (BGP). The classifiers are trained on the feature sets selected by various feature selection algorithms. We compare the Fisher, minimum redundancy maximum relevance (mRMR), extended/weighted/multi-class odds ratio (EORIWORIMOR), and class discriminating measure (CDM) feature selection algorithms. The odds ratio algorithms are extended to include continuous features. The classifiers that are trained based on the features selected by the WOR algorithm achieve the highest F-score.