Agent-Based Modeling for the Analysis of Complex Networks

• Ano de defesa: 2023
• Autor(a) principal: Farfan, Alex Josue Florez
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Agent-based models; Aprendizado de máquina; Complex networks; Machine learning; Modelagem baseada em agentes; Pattern recognition; Reconhecimento de padrões; Redes complexas
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-03042024-113422/

Agent-based modeling is an approach within computational modeling that focuses on simulating the behavior and interactions of individual agents to understand emerging patterns in complex systems. This thesis discusses an approach developed in agent-based models in order to study and analyze complex networks. The inherent characteristics of agent-based models provide the appropriate context for exploring complex networks. By identifying, analyzing and understanding the emergent properties that arise from the dynamics and behavior of the agents we can obtain and recognize patterns within complex networks. Network characterization is an important task of pattern recognition. The modeling of a process over the space provided by networks generate patterns at different levels, individually in the agents, as well as globally in the entire model. In order to achieve the objective, an agent-based approach is proposed from which features are extracted that serve to characterize networks. It is important to highlight that in the literature agent-based models have not been used to categorize networks. The proposed model, called the Growth model, provides a novel consideration to characterize complex networks. The analysis performed on synthetic and real-world network datasets indicate that the classification results are similar with methods of the literature. The classification accuracy shows that in four datasets, Actinobacteria, Fungi, Kingdom, and Plant the results are better than the previous work in the literature, demonstrating the potential of this approach.