Uma arquitetura de referência para explicabilidade como serviço na saúde H-XAIaaS: Health - eXplainable Artificial Intelligence as a Service

• Ano de defesa: 2024
• Autor(a) principal: Montenegro, Thiago Cunha
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Informática; Programa de Pós-Graduação em Informática; UFPB
• Palavras-chave em Português: Inteligência artificial explicável; Aprendizagem de máquina; Suporte à decisão clínica - Sistemas; Arquitetura de referência; XAI; Sistemas explicáveis na saúde; Explicabilidade; AIaaS; Machine learning; Explainability; Artificial intelligence; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/37085

In recent decades, Artificial Intelligence (AI) through the latest Machine Learning tech- niques has impacted various fields, especially healthcare, due to its growing precision and efficiency. Predictive Machine Learning models are promising, but it is crucial that they are understandable and explainable to healthcare professionals, increasing their acceptance and trust. With the advancement of cloud tools, the offering of AI as a service (AIaaS) has grown, allowing organizations and ecosystems to benefit from machine learning in developing solutions. However, the ease of creating and deploying these cloud-based Machine Learning models raises issues of transparency and interpretability, particularly in domains sensitive to these aspects, such as healthcare, which demands integrity and compliance with ethical and regulatory principles. This research proposes a reference architecture based on the AIaaS paradigm to build cloud-based machine learning models, integrating explainability to improve clinical decision-making. The H-XAIaaS architecture aims to propose a reference architecture to enable explainability in AI as a service. To make the proposed architecture feasible, two case studies were implemented. The first case study involved tabular data, and the explainability techniques LIME, Anchor Rules, and counterfactual were used. The second case study involved image data and employed the LIME and Grad-CAM techniques. The proposed architecture proves promising in the con- text of creating machine learning models and their explanations concerning transparency and interpretability, which supports clinical decision-making processes.