Change Detection in the Complexity of Time Series with Information-based Criteria

The objective of the work is to emphasize the need for a complexity measures or coefficients, in time series analysis. Such measures must consider the structure of the model of the signal and the computational effort to process him (i.e. analysis, processing, transmitting, storing, etc). The final goal is to build an automated recognition and classification system for complex signals, i.e. mixtures of various types. In this work the complexity is measured with Renyi entropy of high order, i.e. from 5 to 10. Three types of basic signals are considered, in an independent or mixed approaches: determinist, random and chaotic. Mixed approaches means a weighted summation of the basic components. Change detection in the structure of the time series is of equal importance, because the change is an effect of changing in the structure of the model or system, which have generated the time series. Change detection is based on fusion of some criteria based on information extraction and processing. The results of the experiments indicate the feasibility of the method to discriminate the change in the complexity of the time series.