Gaussian Mixture Model Based Machine Learning Approach for Detection of Threat Types in Communication Networks

The Gaussian mixture model (GMM) based machine learning approach for automatic unsupervised detection of threat types in communication networks is proposed in the paper. The proposed approach uses the expectation-maximization algorithm and a vector of threat associated parameters, which are monitored parameters in communication network. These parameters can be associated with both features of the functioning of communication network and threat scenarios, e.g., delays and number of requests in cases of either specific routing or threats. The detection of threat types in the proposed approach is performed by a forming of subsets of threat associated parameters using latent variables of GMM, which are by nature posterior probabilities that threat associated parameters are associated with components of the GMM. An example of simulation of the proposed approach, which deals with a detection of possible 'no threats', 'insignificant threat', 'moderate threat', and 'significant threat' cases in communication network is shown and analyzed. Features and prospects of the proposed approach are also shown and analyzed in the paper.