Anomaly detection techniques for streaming data–An overview

Abstract

With the advent of smart devices and the Internet, data is being generated from various sources including mobile phones, sensor networks, telecommunications, satellites, log data, business, health care and many government sectors where the data is likely to arrive with speed. Data which flows continuously with respect to time is called streaming data and detection anomaly in such data in real time is an open challenge. Detecting anomaly in right time facilitates the appropriate control actions for the anomaly in right time. There are several techniques for detecting anomaly. In this paper, an overview of different techniques for detection of anomaly is presented.

Are Concept Drift Detectors Reliable Alarming Systems? - A Comparative Study

“…Predictive algorithms, such as classification algorithms using Machine Learning (ML) on Big Data have seen a significant growth in interest and plenty of real-world applications have been proposed. Examples of those applications are fault detection [36], anomaly detection [26], weather prediction [3], or credit risk prediction [6], where different ML models are constantly evaluated on streaming data. Generally, due to the continuous data flow, data streams are more prone to changes in data distributions over time and, thereby, to concept drift.…”