Test for non-negligible adverse shifts

Vathy M Kamulete
Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence, PMLR 180:959-968, 2022.

Abstract

Statistical tests for dataset shift are susceptible to false alarms: they are sensitive to minor differences when there is in fact adequate sample coverage and predictive performance. We propose instead a framework to detect adverse shifts based on outlier scores, D-SOS for short. D-SOS holds that the new (test) sample is not substantively worse than the reference (training) sample, and not that the two are equal. The key idea is to reduce observations to outlier scores and compare contamination rates at varying weighted thresholds. Users can define what worse means in terms of relevant notions of outlyingness, including proxies for predictive performance. Compared to tests of equal distribution, our approach is uniquely tailored to serve as a robust metric for model monitoring and data validation. We show how versatile and practical D-SOS is on a wide range of real and simulated data.

Cite this Paper

BibTeX
@InProceedings{pmlr-v180-kamulete22a, title = {Test for non-negligible adverse shifts}, author = {Kamulete, Vathy M}, booktitle = {Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence}, pages = {959--968}, year = {2022}, editor = {Cussens, James and Zhang, Kun}, volume = {180}, series = {Proceedings of Machine Learning Research}, month = {01--05 Aug}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v180/kamulete22a/kamulete22a.pdf}, url = {https://proceedings.mlr.press/v180/kamulete22a.html}, abstract = {Statistical tests for dataset shift are susceptible to false alarms: they are sensitive to minor differences when there is in fact adequate sample coverage and predictive performance. We propose instead a framework to detect adverse shifts based on outlier scores, D-SOS for short. D-SOS holds that the new (test) sample is not substantively worse than the reference (training) sample, and not that the two are equal. The key idea is to reduce observations to outlier scores and compare contamination rates at varying weighted thresholds. Users can define what worse means in terms of relevant notions of outlyingness, including proxies for predictive performance. Compared to tests of equal distribution, our approach is uniquely tailored to serve as a robust metric for model monitoring and data validation. We show how versatile and practical D-SOS is on a wide range of real and simulated data.} }
Endnote
%0 Conference Paper %T Test for non-negligible adverse shifts %A Vathy M Kamulete %B Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2022 %E James Cussens %E Kun Zhang %F pmlr-v180-kamulete22a %I PMLR %P 959--968 %U https://proceedings.mlr.press/v180/kamulete22a.html %V 180 %X Statistical tests for dataset shift are susceptible to false alarms: they are sensitive to minor differences when there is in fact adequate sample coverage and predictive performance. We propose instead a framework to detect adverse shifts based on outlier scores, D-SOS for short. D-SOS holds that the new (test) sample is not substantively worse than the reference (training) sample, and not that the two are equal. The key idea is to reduce observations to outlier scores and compare contamination rates at varying weighted thresholds. Users can define what worse means in terms of relevant notions of outlyingness, including proxies for predictive performance. Compared to tests of equal distribution, our approach is uniquely tailored to serve as a robust metric for model monitoring and data validation. We show how versatile and practical D-SOS is on a wide range of real and simulated data.
APA
Kamulete, V.M.. (2022). Test for non-negligible adverse shifts. Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research 180:959-968 Available from https://proceedings.mlr.press/v180/kamulete22a.html.

Related Material