Benchmark results

The benchmark runs the detectors of PySAD on labeled data and compares their results. Each detector runs once on each dataset, with its default parameters and seed 0. Use the results to find two or three candidates for your data. Then compare these candidates on your own data, as Choose a detector shows.

A multivariate detector takes a vector of features for each point. A univariate detector takes one value for each point. Benchmark setup explains the steps of a run, the metrics, the datasets and the limits of the results.

The results are on these pages:

• Multivariate detectors: the multivariate detectors on 17 ODDS datasets and on 8 real streams with drift.

• Univariate detectors: the univariate detectors on the NAB and UCR series, and the multivariate detectors on the NAB series.

• Calibrators: the alerts of the two probability calibrators.