A semi-supervised approach to detect malicious nodes in OBS network dataset using gaussian mixture model

Abstract

In this study, we have followed a semi-supervised approach for the classification of optical burst switching (OBS) network traffics generated by the network’s edge nodes. We used expectation maximization (EM) technique for Gaussian mixture model (GMM) to obtain a probabilistic classification of the OBS nodes. For this purpose, we used a trustworthy OBS network dataset from UCI machine learning repository. Preprocessing and principal component analysis were applied to the dataset for arranging the data so that GMM can play its role fairly. Only 1% (10 samples) of labeled data from OBS dataset was used to initialize the parameters of GMM and the rest 99% was used for testing performance of the model. We found a maximum accuracy of 69.7% on the test data using just 1% labeled data with the tied covariance type of the constructed GMM. The significance of this result is that it shows the GMM can be used in classification of OBS networks and other similar networks in semi-supervised way when one has very few labeled data and when labeling a huge dataset is not feasible.