Comparative analysis of clustering techniques in network traffic faults classification

Ubiquitous high-speed communication networks play a crucial role in the modern life, demanding the highest level of reliability and availability. Due to the rapid growth of computer networks in terms of size, complexity and heterogeneity, the probability of network faults increases. Manual network administration is hopelessly outdated; complex automated fault diagnosis and management are essential to ensure the provision and maintenance of high quality service in computer networks. Guaranteed Service with higher levels of reliability and availability for real-time applications can be achieved with a systematic approach for real-time classification of network faults, which helps in well-informed (often-automated) decision making. In this paper we discuss three different data mining algorithms as part of the proposed solution for network fault classification: K-Means, Fuzzy C Means, and Expectation Maximization. The proposed approach can help capture abnormal behavior in communication networks, thus paving the way for real-time fault classification and management. We used data sets obtained from a network with heavy and light traffic scenarios in the router and server and built a prototype to demonstrate the network traffic fault classification under given scenarios. Our empirical results reveal that the FCM is more accurate while causing computational overhead. The other two algorithms attain almost the same performance.