Interpretabilidade com agregação de relevância em redes neurais para a predição do absenteísmo

• Ano de defesa: 2022
• Autor(a) principal: Gomes Junior, Julio Marcos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Cornelio Procopio; Brasil; Programa de Pós-Graduação em Informática; UTFPR
• Palavras-chave em Português: Absenteísmo (Trabalho); Aprendizado do computador; Inteligência artificial; Absenteeism (Labor); Machine learning; Artificial intelligence; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO; Ciência da Computação
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/30199

The lack of attendance of employees is called absenteeism and occurs for various reasons, such as vigorous physical activity, advanced age, and high psychological demands at work. Absenteeism affects direct and indirect costs of companies, and may reach 15% of payroll. Therefore, it is fundamental to know its main causes and contribute to control and mitigation strategies. Neural networks have been successfully applied in the classification of several problems, however they are black boxes, since they do not explain which aspects are considered in their decisions. These aspects are important in healthcare applications, in which it is necessary to clearly explain and interpret the results. In this context, this study presents an approach to classify absenteeism with neural networks, Layer-wise Relevance Propagation (LRP) and relevance aggregation to identify the most relevant features and assign relevance scores individually per class and among all classes. The proposed approach was evaluated by considering a widely used dataset as a reference and comparing with existing methods in the literature. The proposed approach presented the highest assertiveness rate among the compared methods, with an average accuracy of 0.83, identifying the most relevant features for absenteeism classification through a relevance score and it was possible to reduce the dataset features by 75% without significant loss in assertiveness rate. Therefore, the results allow the interpretability of the causes of each absenteeism class and the reduction of dimensionality of the feature space, which contribute to the management of human resources, occupational medicine and the development of strategies for its mitigation.