Application of Explainable Artificial Intelligence (XAI) to detect anomalies in computer networks through the creation of images from data packets
Carregando...
Data
Orientador
Costa, Kelton Augusto Pontara da 

Coorientador
Pós-graduação
Ciência da Computação - FC/FCT/IBILCE/IGCE
Curso de graduação
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Tipo
Tese de doutorado
Direito de acesso
Acesso aberto

Resumo
Resumo (português)
Com a crescente complexidade das ameaças cibernéticas, compreender como os modelos de inteligência artificial tomam decisões tornou-se um passo crucial para melhorar a confiança e a transparência. Este trabalho apresenta uma metodologia que transforma o tráfego de rede em imagens e utiliza Redes Neurais Convolucionais (CNNs) para classificar fluxos como benignos ou maliciosos. Além de alcançar alta precisão preditiva, o estudo integra a técnica de IA Explicável Grad-CAM para explicar visualmente o funcionamento interno do modelo. Essa abordagem permite que os analistas inspecionem quais partes da imagem foram mais influentes na decisão do modelo, facilitando a supervisão humana e a auditoria do sistema. Os resultados destacam o potencial da combinação de aprendizado profundo com ferramentas de interpretabilidade no campo da segurança cibernética.
Resumo (inglês)
With the increasing complexity of cyber threats, understanding how artificial intelligence models make decisions has become a crucial step in improving trust and transparency. This work presents a methodology that transforms network traffic into images and uses Convolutional Neural Networks (CNNs) to classify flows as benign or malicious. Beyond achieving high predictive accuracy, the study integrates the Explainable AI technique Grad-CAM to visually explain the internal workings of the model. This approach allows analysts to inspect which parts of the image were most influential in the model’s decision, facilitating human oversight and system auditing. The findings highlight the potential of combining deep learning with interpretability tools in the field of cybersecurity.
Descrição
Palavras-chave
Network anomalies, Deep learning, Grad-CAM, Explainability, Convolutional neural networks, Anomalias de rede, Aprendizado profundo, Redes neurais convolucionais, Explicabilidade
Idioma
Inglês
Citação
HERNANDES JUNIOR, Paulo Roberto Galego. Application of Explainable Artificial Intelligence (XAI) to detect anomalies in computer networks through the creation of images from data packets. 2025. Thesis (Doctor in Computer Science) – School of Sciences, São Paulo State University (Unesp), Bauru, 2025.


