Abstract

For many real applications, ranging from traditional management information systems (MIS) to e-commerce and socials web, produce large volumes of sequences of multivariate time-stamped observations. The classical data representation for these applications is an information table stored in RDBMS 1, in which rows stand for objects and columns denote attributes. Based on the above scenarios, unsupervised subsequence outlier detection on time series data is a valuable problem in practice which helps in saving the cost of labeling and providing interpretability in real applications. This problem is called subsequence outlier detection tasks [2]. In this abstract, we study this issue and show our preliminary analysis and ongoing work.