Time Series Pattern Discovery by Segmental Gaussian Models

Abstract

As a result of diversification of sensor data due to advances in sensing technology in recent years, large amounts of multidimensional sensor data are stored in various areas such as plants and social systems. It is difficult to take the first step in time series analysis to visualize such sensor data in its entirety. Reflecting the increasing need to analyze data whose features are not clearly understood, the time series analysis method using the features of an economic time series (e.g., ARMA) cannot necessarily be applied. Therefore, methods for analyzing time series data without assuming features of the data are of great interest. The method for extracting features of time series data without assuming features of the data is a time series pattern discovery method [3]. A time series pattern discovery method is used to find the waveforms automatically as time series patterns that arise frequently from time series data.