Publicação
Intrusion detection system, a decentralized approach using multi-agent systems and machine learning
| Resumo: | The increasing adoption of cloud computing and digital environments has made network infrastructures more complex and, at the same time, more vulnerable to cyberattacks. Traditional centralization of security mechanisms is becoming increasingly challenging, since data and applications are dispersed across multiple environments. Faced with this scenario, new decentralized approaches are emerging to try to solve this problem. This work presents a distributed approach for an intrusion detection system, through a multi-agent system and machine learning, and how this can be an advantageous tool compared to a centralized and individual system. Through a multi-agent system, it is possible to solve complex problems collaboratively, using the interaction among multiple autonomous entities that work in a coordinated manner. Each agent within the system has specific objectives, the ability to perceive the environment and the ability to make decisions independently or in conjunction with other agents. This approach is especially useful in dynamic and distributed scenarios, such as logistics, process optimization and large-scale simulations, etc. In addition, multi-agent systems favor the adaptability, scalability and robustness of solutions, allowing an efficient response to challenges that require decentralization and cooperation. Combining this with machine learning techniques, it is possible to add new features to an intrusion detection system, making it intelligent and decentralized, in order to increase the scope of detections, reduce data processing time and reaction. Among the results obtained, comparing a collaborative system with an individual system, accuracy improved by 1.13%, precision by 1.31%, recall (sensitivity) by 0.04% and F1-score by 0.69%. Having an average action time of 6.33 seconds. |
|---|---|
| Autores principais: | Castro, Nathan Cesa Nery De |
| Assunto: | Cybersecurity Intrusion detection system Machine learning Multi-agent systems |
| Ano: | 2025 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Politécnica de Bragança |
| Idioma: | inglês |
| Origem: | Biblioteca Digital da UPB |
| Resumo: | The increasing adoption of cloud computing and digital environments has made network infrastructures more complex and, at the same time, more vulnerable to cyberattacks. Traditional centralization of security mechanisms is becoming increasingly challenging, since data and applications are dispersed across multiple environments. Faced with this scenario, new decentralized approaches are emerging to try to solve this problem. This work presents a distributed approach for an intrusion detection system, through a multi-agent system and machine learning, and how this can be an advantageous tool compared to a centralized and individual system. Through a multi-agent system, it is possible to solve complex problems collaboratively, using the interaction among multiple autonomous entities that work in a coordinated manner. Each agent within the system has specific objectives, the ability to perceive the environment and the ability to make decisions independently or in conjunction with other agents. This approach is especially useful in dynamic and distributed scenarios, such as logistics, process optimization and large-scale simulations, etc. In addition, multi-agent systems favor the adaptability, scalability and robustness of solutions, allowing an efficient response to challenges that require decentralization and cooperation. Combining this with machine learning techniques, it is possible to add new features to an intrusion detection system, making it intelligent and decentralized, in order to increase the scope of detections, reduce data processing time and reaction. Among the results obtained, comparing a collaborative system with an individual system, accuracy improved by 1.13%, precision by 1.31%, recall (sensitivity) by 0.04% and F1-score by 0.69%. Having an average action time of 6.33 seconds. |
|---|
Atividades financiadas
Não foram encontrados projetos financiados.