FederatedLearningOutlierDetection

How to run

TODO

Note

You need a license of gurobi to run the experiments.

Baseline Results

My algorithms are BSVClassifier and OneClassSVM. Both were run 10 times as it can be seen in the code (commit 3bcddb0)

Results Folder

AUC

Classifer	pen-global	breast-cancer	letter	pen-local	annthyroid	satellite	shuttle	aloi	kd99
oc-SVM	0.9512 ± 0.0436	0.9721 ± 0.0102	0.5195 ± 0.0382	0.9543 ± 0.0130	0.5316 ± 0.0152	0.9549 ± 0.0021	0.9862 ± 0.0002	0.5319 ± 0.0021	0.9518 ± 0.0050
OneClassSVM	0.9494 ± 0.0391	0.9793 ± 0.0012	0.6585 ± 0.0798	0.7144 ± 0.2516	0.4933 ± 0.0137	0.9137 ± 0.0284	0.9865 ± 0.0003
BSVClassifier (aka SVDD)	0.9996 ± 0.0009	0.9955 ± 0.0092	0.8080 ± 0.1360	0.7445 ± 0.1723	0.5147 ± 0.0361	0.9395 ± 0.0054

Average Precision

Classifier	pen-global	breast-cancer	letter	pen-local	annthyroid	satellite	shuttle
OneClassSVM	0.9931 ± 0.0059	0.9995 ± 0.0000	0.9601 ± 0.0135	0.9992 ± 0.0010	0.9646 ± 0.0013	0.9980 ± 0.0007	0.9996 ± 0.0000
BSVClassifier (aka SVDD)	0.9999 ± 0.0001	0.9999 ± 0.0003	0.9746 ± 0.0170	0.9994 ± 0.0007	0.9673 ± 0.0038	0.9989 ± 0.0001