Identifying Patterns in Multiple Time Series Data

Abstract

In trying to identify patterns of discrete independent variables in a mass of Multivariate data, J G. Crookes, K. Croston and P. r. Sypsas [5J introduced Process Components Analysis seeking for components which carry the pattern rather than those carrying the variance of the system.The methodology has been developed by P.T. Sypsas [9] and its efficiency has been tested on a number of types of series.An outline of the method and the relevant algorithm is presented here along with a worked example for simulated series of known patterns.The results appear at the end in the form of correlations between the simulated series, used as input to the system, and the resulting process components using the proposed technique.There are also presented graphs of input and output series.