Object

Title in english:

Uczciwość w uczeniu maszynowym – identyfikacja i redukcja uprzedzeń

Creator:

Zaniewska, Aleksandra

Contributor:

Dudycz, Helena. Redaktor

Description:

Informatyka w biznesie / pod red. Heleny Dudycz. - Wrocław: Wydawnictwo Uniwersytetu Ekonomicznego we Wrocławiu, 2022, s. 104-116

Abstrakt:

The author investigates the problem of biases occuring in machine learning algorithms, and the strategies for their identification and mitigation. The biases are classified into three main categories: bias in data, bias in algorithms and bias generated by users. The German credit data set used in this article comes from the UCL Machine Learning Repository and represents credit risk assigned to the applicants applying for credit from the bank. The two machine learning algorithms: Random Forest and XGBoost are trained on this data set, and they are then analysed for the presence of gender bias. Subsequently, pre-processing mitigation bias techniques are used to minimize the impact of gender bias. It is identified that both algorithms have bias present and the False Negative Rate for females is the most common problem. The mitigation strategies help reduce bias but do not reduce them completely.

Publisher:

Wydawnictwo Uniwersytetu Ekonomicznego we Wrocławiu

Place of publication:

Wrocław

Date:

2022

Resource Type:

rozdział

Resource Identifier:

oai:dbc.wroc.pl:118541

Language:

eng

Relation:

Debiuty Studenckie 2022 ; Informatyka w biznesie

Rights:

Pewne prawa zastrzeżone na rzecz Autorów i Wydawcy

Access Rights:

Dla wszystkich zgodnie z licencją

License:

CC BY-SA 4.0

Location:

Uniwersytet Ekonomiczny we Wrocławiu