Detecting temporal patterns using Reconstructed Phase Space and Support Vector Machine in the dynamic data system

In this paper we present a method for detecting dynamic temporal patterns that are characteristic and predictive of significant events in a dynamic data system. We employ the Gaussian Mixture Model (GMM) to cluster the data sequence into three categories of signals, e.g. normal, patterns and events. The data sequence is then embedded into a Reconstructed Phase Space (RPS) which is topologically equivalent to the dynamics of the original system. We apply a hybrid method using Support Vector Machines (SVM) and Maximum a Posterior (MAP) to classify temporal pattern signals based on the event function. We performed two experimental applications using chaotic time series and Sludge Volume Index (SVI) series related to the Sludge Bulking problem. The proposed hybrid GMM-SVM phase space approach effectively detects temporal patterns and achieves higher predictive accuracy compared with the original RPS framework.