Anomaly Detector

• Unsupervised algorithm, no need for labeled data
• Algorithm is trained on normal state data, then used in unseen data to detect deviations
• Applicable to streaming and offline datasets

Input(s):

• Regression datasets

Main feature(s):

• Multi-variate algorithm that detects sudden or instantaneous deviations from a normal state
• Monitors the health or performance of a device

Output(s):

• Anomaly score between 0 and 1, indicating how likely (1) or unlikely (0) the observed data point is an anomaly

The anomaly detector is part of the Ekkono SDK:

• Modeling: .NET 2.0+ or Python 3.7+ (Windows, macOS, or Linux)
• Deployment: C++17 (or C++14 with included MPark.Variant), delivered as source to be compiled by the customer
• Integration will be done through Ekkono’s C++ API of the compiled library

Data Scientists, machine learning engineers, and software developers that want to run machine learning and anomaly detection on any type of device.

Commercial license