The Anomaly Detection Extension comprises the most well know unsupervised anomaly detection algorithms, assigning individual anomaly scores to data rows of example sets

=== 4.1.1 ===

• Histogram based Outlier Score is now returning the colored table again (as in prior versions).

=== 4.1.0 ===

• Raised the requierement to for 4.1 to RM Version 10.0
• Added a new method "Percentile Threshold" to Detect Outliers Univariate, which gives you the distance to the kth and 1-kth percentile.
• Adapted Detect Outlier (rPCA) so that it works with java11 and RM X

=== 4.0.1 ===

• Fixed a bug where cluster models may create wrong results in applications, since they don't take the cluster sizes from training.
• Fixed a bug where Isolation Forests were not able to run on values <0.
• Fixed a bug where Isolation Forests were not storable if they were grouped.

=== 4.0.0 ===

Please be aware that this release is not backwards compatible, since attributes got new names!

• Anomaly models are now serialized using JSON, not java serialization
• Normal Anomaly models are now IOTablePredictionModels, just like any other learner (except clustering)
• Since Anomaly models are now IOTablePredictionModels their response is prediction, not a score
• Univariate models are still Preprocessing models, but their main score is called prediction. all other scores are called prediction(attributeName)
• Univariate models are now stating the correct covered attributes in their description

=== 3.3.0 ===

Warning: This version is not backwards compatible, since column names changed from v 3.2.0 to 3.3.0 to have one scheme.

•  Changed the role of outlier scores in all Detect Outlier operators. Its now a confidence (score).
• Changed the name of the generated score attribute in all Detect Outlier operators. It is now "score"
• Added a new operator "Generate Outlier Flag" which allows you to discretize the score attributes.

=== 3.2.0===

* Added a new operator Detect Outliers (Isolation Forest), which was previously part of Operator Toolbox

* Trees in isolation forests are now trained on a bootstrapped subset of the original data

* Trees in isolation forests can now define the number of features considered in every tree.

=== 3.1.0 ===

* Added a new operator Detect Outliers (Univariate), which was previously part of Operator Toolbox

* Added a new operator Detect Outliers (Time Series)

=== Version 3.0.1 ===

• Added new operator Detect Outliers (Clustering) which wraps the three outlier detection algorithms
• Detect Outliers (Clustering) provides a model to be applied on new data.
• Added a new operator Detect Outliers (rPCA), which also provides a model.
• Switched to an up-to date LibSVM version. Robust one-class is thus not available anymore. Results may differ slightly.
• Switched the lib for RNN operator, results may differ.

=== Version 3.0.2 ===

• Fixed minor bugs