Current Trends in Time Series Representation

Abstract

Time series data generation has been exploded in almost every domain such as in business, industry, medicine, science or entertainment. Consequently, there is an increasing need for analysing efficiently the huge amount of this information either online or offline. The inherent characteristics of time series data, specifically, the high dimensionality, the high feature correlation and the large amounts of noise led researchers to focus mostly on proposing novel representation schemes for this type of data in order to address the resulting problems. In addition to that, representation schemes have also been proposed for the purpose of exploiting specific methods and algorithms that require different type of data, such as Markov models. Concurrently, since most of the data mining tasks require searching for similar patterns, such as query by content, clustering or classification, many papers have been published that propose a generic or representation-based similarity measure. The objective of this paper is to serve as an overview of the recent approaches and trends in the field of time series data mining representations. It seems that the main interest has been gradually drawn in manipulating streaming data and / or multivariate time series. Although a general overview is included, the literature review is focused mainly on papers of the last four years.