Everything You Need to Know About Numenta Anomaly Benchmark (NAB)

Everything You Need to Know About Numenta Anomaly Benchmark (NAB)

Detecting anomalies in datasets has broad applications in a lot of domains such as fraud detection for credit cards, fault detection in safety critical systems, and network intrusion detection for cyber security. Due to its wide range of applications, the problem of anomaly detection has been thoroughly researched by industry specialists and academia alike, and several algorithms have been introduced for determining anomalies in problem settings. However, there wasn’t any dataset or framework openly available using which the proposed algorithms could be evaluated on a common ground.

The situation changed with the introduction of the Numenta Anomaly Benchmark (NAB) by Numenta in 2015. It is an open source framework used to evaluate and compare various anomaly detection algorithms.

Numenta Anomaly Benchmark (NAB)

As mentioned above, the Numenta Anomaly Benchmark (NAB) is a set of openly-available, labeled data files and common scoring system to compare and evaluate different anomaly detection algorithms for detecting anomalies in streaming data. Anomalies (also referred to as outliers) in streaming data are patterns that do not match with previous patterns of behavior for the given data stream.