Highlight which attributes are outliers within the entire line

Jeffersonjpa

In the process of detecting outliers how do I know / highlight which attributes are outliers. Example: I know I can use the Detect Outlier (LOF)  component but it highlights the entire line as outliner. How do I know within the line which of the attributes is really nonstandard ... which of the attributes are really outliers?

User146561

Outliers are typically determined by a deviation from the norm in a high dimensional space. That means, there is not a single attribute responsible for an outlier but a combination thereof. Short answer to your question: Not possible.

However, there is one algorithm which assumes independence of the featues and each feature contributes to an outlier score. The algorithm is called HBOS and maybe what you are looking for. You can find it in the "Anomaly Detection Extension". There is also a visualization as a colored table for the results illustrating which feature of which row causes a high outlier score.

tftemme

Hi @Jeffersonjpa,

In addition to the solution from @User146561 you can use the Tukey Test operator from the Operator Toolbox extension. This operator only calculates attribute specific outlier scores. Obviously it does not take into account any dependencies between the attributes to determine the outlier score for each example. But you maybe can combine both and check for Examples with a high LOF-score, what the score of the different tukey tests are. This is of course only a hint, which attributes may contribute to the LOF-Score.

Best regards,
Fabian