Título: Deep Reinforcement Learning Using Double Deep Q-Network Applied to Anomaly Detection
Autores: João Oldenburg, Enio dos Santos Silva
Resumo: The digital transformation and the exponential growth of data generated daily have driven a global data-driven landscape, where companies increasingly seek to extract strategic value from large datasets. In this context, machine learning techniques and neural networks have become essential tools for efficiently processing and interpreting data. Anomaly detection, crucial for identifying unusual patterns, plays a fundamental role and is a highly relevant research area. To investigate and identify anomalous behaviors in time series, this work explores the application of deep neural networks combined with reinforcement learning techniques. Among the studied approaches, the Double Deep Q-Network (DDQN) stands out as an effective method in scenarios requiring the generalization of large datasets. Using the “Numenta Anomaly Benchmark” (NAB) dataset, this study investigates the effectiveness of DDQN for anomaly detection. The results highlight not only the academic contribution of this approach but also its practical impact on developing robust solutions for critical problems in dynamic and complex environments.
Palavras-chave: Deep reinforcement learning; anomaly detection; artificial intelligence; deep neural networks; time series.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1175438
Artigo em PDF: CBIC_2025_paper1175438.pdf
Arquivo BibTeX:
CBIC_2025_1175438.bib