Clasificación jerárquica semántica aplicada a la detección de anomalías mediante logs de sistemas
Year: 2024
Type: dissertation
Abstract: [Abstract] The objective of this study focuses on the preprocessing of system logs to make them more compact and structured, thus facilitating centralized processing that uses machine learning techniques to detect anomalies. This thesis implements an algorithm that creates and uses codes from system log themes to generate a log compendium that maintains semantics, thus reducing processing in anomaly detection. Develops a hierarchical classification of themes of natural language words and codes each theme with hierarchical codes that represent the specificity of the theme in the hierarchy. Creates an alphabetical catalog of words that may exist in the logs associated with their hierarchical codes. This catalog is used to analyze and reduce the size of Hadoop Distributed File System (HDFS) logs, obtaining reduced logs without repeated themes and thus improving efficiency in automatic anomaly detection analysis. The log compendium has been validated for anomaly detection using the “bert-base-uncased” model and has been compared with LogCluster, SVM, LogRobust, DeepLog, PCA and IM, obtaining much shorter processing times, but similar values in precision, recall and F1-score metrics.
Source: RUC (Universidade Da Coruña)
Author: Corbelle Suárez, Clara
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7027384993