ELKI, exit: ROCAUC, Precision @k, f1.maximum
What is the Precision @k used in the case of a blowout? (when changing k on one dataset I always get: Precision @ 3016 and I don't get where ELKI has this number, the number of outputs is 1508)
and
precision.average and precision.r?
and f1.maximum?
I know that ROCAUC is a measure of how much the algorithm calls outliers as outliers and normal as normal objects.
I want to know if the quality of outlier detection is good. Can I do this with other measures as well?
Computing LDOFs
LDOF for objects: 49534 [100%]
de.lmu.ifi.dbs.elki.algorithm.outlier.lof.LDOF.runtime: 116887 ms
Evaluating using minority class: yes
de.lmu.ifi.dbs.elki.evaluation.outlier.OutlierRankingEvaluation.rocauc: 0.736341684836717
de.lmu.ifi.dbs.elki.evaluation.outlier.OutlierRankingEvaluation.precision.average: 0.10795456476088741
de.lmu.ifi.dbs.elki.evaluation.outlier.OutlierRankingEvaluation.precision.r: 0.16578249336870027
de.lmu.ifi.dbs.elki.evaluation.outlier.OutlierRankingEvaluation.f1.maximum: 0.18336314847942753
ROCAUC: 0.7363416848367167
Precision @ 3016 0.13726790450928383
source to share
R-precision - Precision @k with k = number of outliers in your truth. For now, remember = accuracy. So in your case it would be @ 1508 precision. The "R" comes from the point "recall = precision".
This is not a parameter k of the algorithm. But both are commonly referred to as k in the literature.
The maximum F1 value is the maximum F1 value (precision and repeatability) obtained by changing the cutoff k. So there is a k where the harmonic mean of precision and recall is 0.18336.
Medium Accuracy - Normal Medium Accuracy; estimated at each outlier, then averaged.
For more details see for example https://en.wikipedia.org/wiki/Information_retrieval
Precision @ 2k is an artifact: the automatic estimation will create an @k precision curve up to 2k (usually an area of interest). The curve will output the precision at the end of the curve; but this is generally not particularly useful (you can manually adjust the maximum k for a curve clip, but that doesn't make it an interesting k to judge). I will remove this from the classes; and also a second ROCAUC output is not needed (also comes from the visual curve evaluation).
source to share