Comparative study of unsupervised anomaly detection in sensor data†
Geonwoo Ko1 · Bohyeon Cho2 · Youngju Byun3 · Donghyeon Yu4
123Department of Statistics, Inha University
Correspondence to: † This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT)(NRF-2022R1A5A7033499).
1 Master student, Department of Statistics, Inha University, Incheon 22212, Korea.
2 Master student, Department of Statistics, Inha University, Incheon 22212, Korea.
3 Master student, Department of Statistics, Inha University, Incheon 22212, Korea.
4 Associate professor, Department of Statistics, Inha University, Incheon 22212, Korea. E-mail: dyu@inha.ac.kr
1 Master student, Department of Statistics, Inha University, Incheon 22212, Korea.
2 Master student, Department of Statistics, Inha University, Incheon 22212, Korea.
3 Master student, Department of Statistics, Inha University, Incheon 22212, Korea.
4 Associate professor, Department of Statistics, Inha University, Incheon 22212, Korea. E-mail: dyu@inha.ac.kr
Received May 19, 2023; Revised June 15, 2023; Accepted June 22, 2023.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
- Abstract
- Unsupervised learning-based anomaly detection is one of the active research topics and is widely applied when the training data only consists of normal observations. In this paper, we introduce several anomaly detection methods based on unsupervised machine learning and deep learning models and provide their performance comparison results using frequency-based sensor data. We consider leak detection sensor data available from the AI-Hub platform. To select the optimal tuning parameters, we consider the self-adaptive data shifting method that generates pseudo anomaly observations and pseudo normal observations from the normal observations. From our comparison results, the DASVDD model that combines the autoencoder and Deep SVDD models showed the best anomaly detection performance in leak detection sensor data.
- Keywords : Anomaly detection, deep learning models, machine learning models, sensor data, unsupervised learning
Full Text(PDF) Free
