A Correlation and Clustering Algorithm for Time Series Monitoring of Network Flows

Over the last few decades the advent of machine learning has found applications in large number of domains. The ability to learn and identify patterns in data and make predictions on deviations from these patterns has found large scope in several fields. In computer networks, this problem can be applied to understanding the behavior of a network at any point of time and predicting when the behavior may change over time. This paper discusses an approach that uses a statistical machine learning algorithm for the time series behavior analysis of computer networks. The proposed algorithm makes use of unsupervised learning and statistical data analysis methods over the flows in the network. The novel aspect being explored is the analysis of the inter-dependencies between flows, in addition to monitoring anomalies of individual flows. The results presented, provide waveform representations of the correlation and clustering patterns of some sample flows based on packet sizes.