Frequent pattern mining from multivariate time series data

Abstract

Discrete sequential data is the collection of an ordered series of discrete events or abstractions from a set of data points collected at different time periods. These processes are one of the most common and important process types encountered in various domains. It is customary to discover similarities and to detect the indicators of anomalies in multivariate form in a supervised setting. In this paper, we first use an effective data transformation technique that transforms multivariate time series into multivariate sequences and use a tree-based method to mine frequent patterns from multivariate time series. However, this problem is costly in terms of solution time and memory consumption. Specifically, this study aims to improve computational efficiency for memory with reasonable solution time. We demonstrate the efficiency of this algorithm on standard datasets and then apply the method to a real healthcare problem.