What is Anomaly Detection?

Anomaly Detection is the identification of rare items, events, or observations in data that differ significantly from the expectation.

Anomaly Detection service is designed to help with analyzing large amounts of correlated data and identifying the anomalies at the earliest possible time with maximum accuracy. Use cases of anomaly detection are in various markets include:

  • Utility

  • Oil and Gas

  • Transportation

  • Manufacturing

  • Telecommunications

  • Banking

  • Insurance

  • Web Businesses

  • E-commerce

In each of these verticals, the anomaly detection service can be used to identify undesirable business incidents and observations, provide the magnitude of anomaly as the difference between expected and actual value. This service helps to define business-specific alerts and actions. It helps you to identify anomalies in a multivariate dataset by taking advantage of interrelationship between signals. When the service builds a machine learning model for each of the signals as a function interrelationship between signals, it maximizes accuracy of the identified anomalies. This accuracy helps to monitor complex systems with large number of signals.

The anomaly detection service utilizes an innovative statistical method that helps to identify anomalies at the earliest possible time. Also, it productizes the Oracle Multivariate State Estimation Technique (MSET) with Sequential Probability Ratio Test (SPRT).

Key Terms

Review these terms used by the core Machine Learning (ML) engine in the Anomaly Detection service:

Technique

Description

Anomaly Detection

Multivariate State Estimation Technique (MSET) An advanced pattern recognition method that learns the correlation between multiple signals over a large dataset of time series data. It provides accurate estimates for a given timestamp.
Sequential Probability Ratio Test (SPRT) A method that takes the estimates generated by MSET and compares them against the original signal value at a particular timestamp to decide whether the signal value is an anomaly.

Intelligent Data Preprocessing (IDP) techniques

A combination of different data preprocessing techniques to resolve different data quality issues from the data before training the model. IDP includes the 3 methods defined in the 3 rows. For example, ARP, MVI, and UNQ
Analytical Resampling Process (ARP) Used for aligning the signal values in a time series dataset with multiple signals that emit data that are not synchronized. Commonly used when there are clock out of sync problems.
Missing Value Imputation (MVI) Used for deriving the missing sensor data in a dataset. Commonly used when signals are not reported because of component failures.
UnQuantization (UnQ) Used for improving the quality of low-resolution input signals to a higher resolution. Commonly used in IoT applications where sensors send low-resolution signals.