Compression of Cyclic Time Series Data
Source: Tampere University of Technology
This paper describes a method to reduce the space of stored indicator time series data collected from a telecommunications network. The method takes advantage of the cyclic nature of the collected time series and behavioral clusters of the network elements. The method separates the shape of the daily activity cycle of a network element from volume of the traffic and clusters the daily patterns to coherent groups in which each sample can be presented with the amount of traffic in the network element and a prototype pattern. Dynamic thresholds are used to determine the samples that deviate from the prototype too much and require being stored separately.