ANN: MultivariateAnomalies.jl - package for detecting multivariate anomalies or novelty detection



I’m happy to announce, that we finally released MultivariateAnomalies.jl - A julia package for detecting multivariate anomalies.

Keywords: Novelty detection, Anomaly Detection, Outlier Detection, Statistical Process Control, Process Monitoring

Currently implemented are some anomaly detection algorithms, specifically:

  • Kernel density estimation,
  • two k-nearest neighbours approaches,
  • Hotelling’s T^2,
  • a recurrence approach,
  • Support vector data description and
  • Kernel Null Foley Summon Transform

A comparison of the implemented algorithms on Earth observation like data can be found here: doi:10.5194/esd-8-677-2017

Package Features

  • Detect anomalies in your data with easy to use high level functions or individual anomaly detection algorithms.

  • Feature Extraction: Preprocess your data by extracting relevant features

  • Postprocessing: Postprocess your anomaly scores, by computing their quantiles or combinations of several algorithms (ensembles).

  • AUC: Compute the area under the curve as external evaluation metric of your anomaly scores.

  • Online Algorithms: Selected anomaly detection algorithms tuned for little memory allocation (to be extended).


The Documentation is available here