Stage-oe-small.jpg

Inproceedings3852

Aus Aifbportal
Version vom 11. Dezember 2020, 10:51 Uhr von Yj7609 (Diskussion | Beiträge) (Die Seite wurde neu angelegt: „{{Publikation Erster Autor |ErsterAutorNachname=Rädsch |ErsterAutorVorname=Tim }} {{Publikation Author |Rank=2 |Author=Sven Eckhardt }} {{Publikation Author |…“)
(Unterschied) ← Nächstältere Version | Aktuelle Version (Unterschied) | Nächstjüngere Version → (Unterschied)
Wechseln zu:Navigation, Suche


What Your Radiologist Might be Missing:Using Machine Learning to Identify Mislabeled Instances of X-ray Images


What Your Radiologist Might be Missing:Using Machine Learning to Identify Mislabeled Instances of X-ray Images



Published: 2021 Januar
Herausgeber: HICSS
Buchtitel: Proceedings of the 54th Hawaii International Conference on System Sciences (HICSS)
Verlag: Springer

Nicht-referierte VeröffentlichungNote: Die Veranstaltung findet wegen der Corona-Pandemie als Online-Event statt.

BibTeX

Kurzfassung
Label quality is an important and common problemin contemporary supervised machine learning research.Mislabeled instances in a data set might not only impactthe performance of machine learning models negativelybut also make it more difficult to explain, and thus trust,the predictions of those models. While extant researchhas especially focused on the ex-ante improvementof label quality by proposing improvements to thelabeling process, more recent research has startedto investigate the use of machine learning-basedapproaches to identify mislabeled instances in trainingdata sets automatically.In this study, we proposea two-staged pipeline for the automatic detection ofpotentially mislabeled instances in a large medical dataset. Our results show that our pipeline successfullydetects mislabeled instances, helping us to identify 7.4%of mislabeled instances of Cardiomegaly in the dataset. With our research, we contribute to ongoing effortsregarding data quality in machine learning.



Forschungsgruppe

Critical Information Infrastructures


Forschungsgebiet