The dynamics of internet of things

• Ano de defesa: 2021
• Autor(a) principal: João Batista Borges Neto
• Orientador(a): Antonio Alfredo Ferreira Loureiro
• Banca de defesa: Alejandro César Frery Orgambide, Osvaldo Anibal Rosso, Thaís Vasconcelos Batista, Pedro Olmo Stancioli Vaz de Melo, Mario Sérgio Ferreira Alvim Júnior
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Minas Gerais
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação
• Departamento: ICEX - INSTITUTO DE CIÊNCIAS EXATAS
• País: Brasil
• Palavras-chave em Português: Internet of Things; Collaborative sensing; Time series dynamics; Ordinal patterns transformations
• Link de acesso: http://hdl.handle.net/1843/41411; https://orcid.org/0000-0001-6497-1613

This thesis investigates the dynamical behavior of time series data from Internet of Things (IoT) sensors. Because of the growing number of IoT initiatives, with its impressive number of devices collecting a large amount of data from real-world phenomena, there is an imminent need for solutions that are adequate to their issues. For instance, an important part of the current IoT is the Collaborative Internet of Things (CoIoT), which is mainly composed by cheap components and managed by common users, affecting the generated data. Thus, solutions for IoT must consider improving the security of those devices, as well as the quality and reliability their data, but being efficient and robust to the issues from this novel scenario. A subject that has been successfully used for a deeper comprehension of several real-world phenomena is the study of dynamics, which aims to understand systems that evolve in time. An important tool with solid results concerning the analysis of time series dynamics is the ordinal patterns transformation. However, while the dynamics has the potential to be the basis for novel representation domains to the analysis of IoT data, there are issues on their transformations that must be handled for their proper applicability. This work aims to advance the state-of-the-art in the analysis of time series dynamics, to be adequate for the IoT issues, and to propose solutions based on dynamical behavior for a more reliable use of data from IoT. In order to advance the applicability of ordinal patterns transformations for challenging scenarios, such as IoT, we propose strategies in two main directions. A first strategy is aimed to provide minimum dependency on the selection of parameters by the transformation, by considering the multiscale behavior of a novel proposed metric, the probability of self-transitions, which are shown to be useful for the distinction of time series dynamics. The second strategy consists of a class separability index, which is a valuable method to estimate the most adequate parameters for the ordinal patterns transformations, in the context of time series classification problems. With respect to the application of the analysis of time series dynamics to IoT scenarios, we first give an enlightenment on the CoIoT. We provide a better understanding on the main characteristics and properties of data that are being generated by their sensors and its inherent problems. Then, we provide strategies for the classification of physical phenomena data collected by CoIoT sensors and a method to increase the security of IoT devices against botnet attacks, both considering their dynamical behavior. The proposed strategies were compared to related work and the results show their potentials on advancing the applicability of ordinal patterns transformations for the IoT scenarios. We show that the construction of this novel representation helps in the scalability, avoiding comparisons with a large number of data, and being robust to the problems of CoIoT data. Thus, by following these approaches, it is possible to develop solutions for IoT scenarios that can benefit from the unique aspects of dynamical systems.