Resumen:
This work presents the Névoa das Coisas Confiável (do Inglês Realible Fog of Things) (RFoT), a framework that integrates an Internet das Coisas (do inglês Internet of Things) (IoT) network with Blockchain and Contratos Inteligentes (do Inglês Smart Contracts) (CI) technologies, to offer IoT as a reliable data source for training Aprendizado de Má- quina (do Inglês Machine Learning) (AM) models. The reliability concept applied in this research is grounded in the four pillars of information security: confidentiality, integrity, availability, and authenticity. Confidentiality was subdivided into process automation and privacy, associated with CI and Fernet synchronous cryptography, while integrity and authenticity were linked to the immutability and traceability provided by Block- chain. This reliability concept was evaluated through experiments that subjected the RFoT and a conventional IoT to stress situations. In this research, a conventional IoT is understood as one that does not implement measures to ensure the reliability of col- lected samples. In these experiments, a AM system was used, based on the Aprendizado Federado (do inglês, Federated Learning) (AF) algorithm to train a Rede Neural (do in- glês Neural Network) (RN) capable of predicting the thermal comfort of an environment, guided by the calculation of the Indice de Desconforto Térmico (IDT). The AM system acted as a consumer in the experiments, to validate whether corrupted data is being propagated by the source and its impact on trained models. The results revealed that a conventional IoT propagates corrupted data, which affects the classification capacity of the models.
