Algorithms overview
- Alessandra Cagnazzo
Our machine learning pipeline is composed by a series of algorithms that altogether contribute to build an alert.
Â
Â
The data are streamed live to the ML pipeline, preprocessed, and piped to two different branches. One (in the image the upper branch) that takes care of the univariate anomaly detection, and another (the lower one) that takes care of computing correlations and grouping together different time series according to their correlations. The univariate anomaly detection and the groups based on correlation are then combined into an alert that deliver jointly anomalies happening within the same correlation group.
Our algorithms are learning from data, so expect for the anomaly alerting to be more noisy for the first few weeks, as we gather data.
In the following you will find detailed descriptions of each part of the ML pipeline:
Onboarding, preprocessing and filtering of the data
Correlation and grouping of time series