Abstract
Outlier detection research is currently focusing on the development of new methods and on improving the computation time for these methods. Evaluation however is rather heuristic, often considering just precision in the top k results or using the area under the ROC curve. These evaluation procedures do not allow for assessment of similarity between methods. Judging the similarity of or correlation between two rankings of outlier scores is an important question in itself but it is also an essential step towards meaningfully building outlier detection ensembles, where this aspect has been completely ignored so far. In this study, our generalized view of evaluation methods allows both to evaluate the performance of existing methods as well as to compare different methods w.r.t. their detection performance. Our new evaluation framework takes into consideration the class imbalance problem and offers new insights on similarity and redundancy of existing outlier detection methods. As a result, the design of effective ensemble methods for outlier detection is considerably enhanced.
| Original language | English |
|---|---|
| Title of host publication | Proceedings of the 12th SIAM International Conference on Data Mining |
| Editors | Joydeep Ghosh, Huan Liu, Ian Davidson, Carlotta Domeniconi, Chandrika Kamath |
| Publication date | Dec 2012 |
| Pages | 1047-1058 |
| ISBN (Print) | 9781611972320 |
| ISBN (Electronic) | 978-1-61197-282-5 |
| DOIs | |
| Publication status | Published - Dec 2012 |
| Externally published | Yes |
| Event | 12th SIAM International Conference on Data Mining - Anaheim, United States Duration: 26. Apr 2012 → 28. Apr 2012 |
Conference
| Conference | 12th SIAM International Conference on Data Mining |
|---|---|
| Country | United States |
| City | Anaheim |
| Period | 26/04/2012 → 28/04/2012 |
| Sponsor | American Statistical Association |
Fingerprint
Cite this
}
On evaluation of outlier rankings and outlier scores. / Schubert, Erich; Wojdanowski, Remigius; Zimek, Arthur; Kriegel, Hans Peter.
Proceedings of the 12th SIAM International Conference on Data Mining. ed. / Joydeep Ghosh; Huan Liu; Ian Davidson; Carlotta Domeniconi; Chandrika Kamath. 2012. p. 1047-1058.Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
TY - GEN
T1 - On evaluation of outlier rankings and outlier scores
AU - Schubert, Erich
AU - Wojdanowski, Remigius
AU - Zimek, Arthur
AU - Kriegel, Hans Peter
PY - 2012/12
Y1 - 2012/12
N2 - Outlier detection research is currently focusing on the development of new methods and on improving the computation time for these methods. Evaluation however is rather heuristic, often considering just precision in the top k results or using the area under the ROC curve. These evaluation procedures do not allow for assessment of similarity between methods. Judging the similarity of or correlation between two rankings of outlier scores is an important question in itself but it is also an essential step towards meaningfully building outlier detection ensembles, where this aspect has been completely ignored so far. In this study, our generalized view of evaluation methods allows both to evaluate the performance of existing methods as well as to compare different methods w.r.t. their detection performance. Our new evaluation framework takes into consideration the class imbalance problem and offers new insights on similarity and redundancy of existing outlier detection methods. As a result, the design of effective ensemble methods for outlier detection is considerably enhanced.
AB - Outlier detection research is currently focusing on the development of new methods and on improving the computation time for these methods. Evaluation however is rather heuristic, often considering just precision in the top k results or using the area under the ROC curve. These evaluation procedures do not allow for assessment of similarity between methods. Judging the similarity of or correlation between two rankings of outlier scores is an important question in itself but it is also an essential step towards meaningfully building outlier detection ensembles, where this aspect has been completely ignored so far. In this study, our generalized view of evaluation methods allows both to evaluate the performance of existing methods as well as to compare different methods w.r.t. their detection performance. Our new evaluation framework takes into consideration the class imbalance problem and offers new insights on similarity and redundancy of existing outlier detection methods. As a result, the design of effective ensemble methods for outlier detection is considerably enhanced.
U2 - 10.1137/1.9781611972825.90
DO - 10.1137/1.9781611972825.90
M3 - Article in proceedings
AN - SCOPUS:84874048074
SN - 9781611972320
SP - 1047
EP - 1058
BT - Proceedings of the 12th SIAM International Conference on Data Mining
A2 - Ghosh, Joydeep
A2 - Liu, Huan
A2 - Davidson, Ian
A2 - Domeniconi, Carlotta
A2 - Kamath, Chandrika
ER -