Robust method for anomaly detection and extreme event prediction in time series data

Abstract

Time series data is ubiquitous in real-world problems across various domains including newlinehealthcare, social media, financial transactions, and crime surveillance. Detecting anomalies, or irregular and rare events, in time series data can enable us to find abnormal events in any natural phenomena, which may require special treatment. Anomalies are rare, often irregular but organic events, that may be identified due to their contrasting behavior from the vast majority in the dataset. For example, in financial transaction normal expenditure of a person follows a pattern whereas some fraudulent interruption causes abrupt variation in this behavior, as in the case of astronomical data apart from the normal sky observations it may sometimes include radio/solar burst signals. Anomalies are these extreme events and are not only limited to this. Moreover, labeled instances of anomaly are hard to get in time series data, making unsupervised anomaly detection techniques crucial. In this work, we initially address the problem of anomaly detection in time series data and then focus on extreme event prediction in time series. newline