Item Response Theory in Sample Reweighting to Build Fairer Classifiers

Autor(es)

Minatel, Diego ; dos Santos, Nicolas Roque ; da Silva, Vinicius Ferreira ; Curi, Mariana ; Lopes, Alneu de Andrade

• Tipo de documento: Artigo Científico
• Fonte: INFORMATION MANAGEMENT AND BIG DATA, SIMBIG 2023; v. 2142, p. 15-pg., 2024-01-01.

Resumo

Currently, one of the biggest challenges of Machine Learning (ML) is to develop fairer models that do not propagate prejudices, stereotypes, social inequalities, and other types of discrimination in their decisions. Before ML faced the problem of unfair decision-making, the field of educational testing developed several mathematical tools to decrease bias in selections made by tests. Thus, the Item Response Theory is one of these main tools, and its great power of evaluation helps make fairer selections. Therefore, in this paper, we use the concepts of Item Response Theory to propose a novel sample reweighting method named IRT-SR. The IRT-SR method aims to assign weights to the most important instances to minimize discriminatory effects in binary classification tasks. According to our results, IRT-SR guides classification algorithms to fit fairer models, improving the main group fairness notions such as demographic parity, equal opportunity, and equalized odds without significant performance loss. (AU)

• Processo FAPESP: 20/09835-1 - IARA - Inteligência Artificial Recriando Ambientes
• Beneficiário: André Carlos Ponce de Leon Ferreira de Carvalho
• Modalidade de apoio: Auxílio à Pesquisa - Programa Centros de Pesquisa em Engenharia
• Processo FAPESP: 22/09091-8 - Criminalidade, insegurança e legitimidade: uma abordagem transdisciplinar
• Beneficiário: Luis Gustavo Nonato
• Modalidade de apoio: Auxílio à Pesquisa - Programa eScience e Data Science - Temático