This thesis deals with the problem of anomaly detection for time series data. Some of
the important applications of time series anomaly detection are healthcare, eco-system
disturbances, intrusion detection and aircraft system health management.
Although there has been extensive work on anomaly detection (1), most of the
techniques look for individual objects that are different from normal objects but do not
consider the sequence aspect of the data into consideration.
In this thesis, we analyze the state of the art of time series anomaly detection
techniques and present a survey. We also propose novel anomaly detection techniques
and transformation techniques for the time series data. Through extensive experimental
evaluation of the proposed techniques on the data sets collected across diverse domains,
we conclude that our techniques perform well across many datasets.