A Novel Method for Assessing Event Impacts on Event-Driven Time Series

Abstract

Many real-world applications, such as service execution, data centre monitoring, remote sensing, traffic control, customer behaviour, have to deal with the time series which include the values occurring at random time points driven by events. These kinds of time series are sometimes also referred to as event-driven time series. Although estimating the correlation between two time series has been well studied, the correlation between events and time series has been understudied. This paper introduces a novel method for assessing event impacts on event-driven time series. In this paper, we estimate the actual event impact time on a time series using a novel and generic algorithm SPEAK. Furthermore, we propose a novel metric Ascore to qualitatively and quantitatively measure the event impact. Our experiments on real-world datasets suggest the combination of Ascore and SPEAK achieved much more accurate results compared to benchmarks.