algorithm - Outlier dectection Using ELKI -

algorithm - Outlier dectection Using ELKI -

i utilize elki info mining software outlier detection. have many outliers detection techniques provides same results(same outliers techniques difference in size of circle around points shown in figures below). uses mouse head dataset provided on elki website. in data-set points labeled respective cluster name, whether ear_left or ear_right or head or noise. if alter label of noise ear_right, shows outlier point ear_right. have alter 5 out of 10 noise label ear_right.

here result of using knn , ldof outlier detection technique modified data-set , in elki:

is problem software or doing wrong? have tried using outlier detection? there software can perform outlier detection using different algorithms lof, ldof , knn or find algorithm source code these techniques?

this simplistic info set.

it not surprising methods all work more or less good. because toy info set, not real data... on real data, outlier detection much, much harder.

note implementations in elki assign numerical scores. not produce yes/no outlier decision; trivial derive scores.

if want binary result, can illustration set visualization scaling parameter visualize top k results. in other cases, may want read actual papers. example, authors of loci suggest treat objects score larger 3 outliers. (unfortunately, methods not have particular easy interpretation available.)

don't think in classification box. outlier detection explorative technique, not classification.

elki can evaluate quality of outlier method using number of measures, such roc auc, roc curves, precision@k, avep, maximum-f1.

algorithm data-mining detection outliers elki