Data Drift Detection

Monitor changes in real-world data distributions

When machine learning models are deployed in production, sometimes even minor changes in a data distribution can adversely affect the performance of ML models. When the input data distribution shifts then prediction quality can drop. It is important to track this kind of data drift. Drift detection in Seldon Enterprise Platform is powered by our open source library alibi-detect. Learn more about the data drift detection concepts on the alibi-detect documentation page.

Available Detection Methods

Seldon Enterprise Platform supports a subset of the methods currently available in alibi-detect for Seldon Core deployments - currently only offline drift detection on input data streams to ML models are supported. Input streams for drift detection may use any protocol currently supported by Seldon Enterprise Platform (Seldon V1, TFServing, and the V2 protocol). Seldon Enterprise Platform enables monitoring of drift detection metrics in real-time and on historical data, both at a feature-level and batch-level as per the detection method. Find documentation on the supported alibi-detect drift detection methods in the table below.

Drift detection method	Detection Type	Alibi-Detect docs
Kolmogorov-Smirnov Drift	Offline	Method Docs
ChiSquare Drift	Offline	Method Docs
Maximum Mean Discrepancy Drift	Offline	Method Docs
Tabular Drift	Offline	Method Docs
Classifier Drift	Offline	Method Docs

Demo

Try out the drift detection and distribution comparison demo on mixed-type tabular data in Seldon Enterprise Platform.