Contextual Anomaly Detection in Time Series Using Dynamic Bayesian Network

Abstract

In this paper, we propose a novel method to identify contextual anomaly in time series using Dynamic Bayesian Networks (DBN). DBN is a powerful machine learning approach that captures temporal characteristics of time series data. In order to detect contextual anomaly we integrate contextual information to the DBN framework, referred to as Contextual DBN (CxDBN). The efficacy of CxDBN is shown using a case study of the identification of contextual anomaly in real-time oil well drilling data.