Anomaly Detection in Logs Using Deep Learning

Detection of abnormalities is important for the security and reliability of computer systems as they heavily rely on logs to detect anomalies. The logs provide general information, errors, warnings, and debugging information. Existing approaches for detecting anomalies are sometimes inaccurate due t...

Full description

Saved in:
Bibliographic Details
Main Authors: Ayesha Aziz, Kashif Munir
Format: Article
Language:English
Published: IEEE 2024-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10767232/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Detection of abnormalities is important for the security and reliability of computer systems as they heavily rely on logs to detect anomalies. The logs provide general information, errors, warnings, and debugging information. Existing approaches for detecting anomalies are sometimes inaccurate due to their limitations related to log-processing leading to loss of semantic significance. Existing approaches, like Deeplog and LogAnomaly, have restrictions in detecting irregularities in log frameworks mainly in large dynamic systems. In this paper, we propose a hybrid anomaly detection technique that combines unsupervised approaches such as Self-Organizing Maps, Bert Encoders, and Autoencoders to handle these issues. The approach improves anomaly identification accuracy by employing semantic vectors obtained by the Bert Encoder to recognize patterns with autoencoders. The evaluation results show that the proposed strategy outperforms the existing methods for various types of data including system logs, network traffic, and financial transactions.
ISSN:2169-3536