Multimodal Fusion for System-Wide Anomaly Detection through Multiple Log Files

Bosch, Nathan (2020) Multimodal Fusion for System-Wide Anomaly Detection through Multiple Log Files. Bachelor's Thesis, Artificial Intelligence.

Preview	Text AI_BA_2020_NathanBosch.pdf Download (616kB) | Preview
	Text Toestemming.pdf Restricted to Registered users only Download (92kB)

Abstract

Complex software-intensive systems write information about their runtime behavior into log files, which are frequently used for both post-mortem and real-time analyses. These analyses can be used to determine anomalous occurrences in the log files. Machine learning techniques have previously been used to automate these analyses on single log files. However, as complex systems often produce many log files, each representative of some differing abstraction level or subsystem, it is difficult to use the existing techniques for system-wide anomaly detection. In this thesis, we aim to fill this gap by approaching multi-log anomaly detection from a multimodal perspective, designing and evaluating two early fusion models and one late fusion model to detect anomalies in log files. The models were evaluated on internal base station test data at Ericsson AB and, for the presentation of results, on open-source log file data. On the open-source data, the early fusion models both achieved an F1-Score greater than 0.95 when detecting system anomalies, whereas the late fusion model achieved an F1-Score of 0.88. However, we found that the performance of the models is dependent on the dependencies and relations between the log data, indicating that different fusion strategies may perform better in different situations.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Schomaker, L.R.B.
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	23 Jul 2020 11:27
Last Modified:	14 Aug 2020 13:15
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/22523

Actions (login required)

View Item