Comparative Analysis of Anomaly Detection Models using AE and Suggestion of Criteria for Determining Outliers (오토인코더를 사용한 이상탐지 모델의 비교분석 및 이상치 판별 기준 제안)

Gun-Ha Kang (강건하); sohn jung mo (손정모); Gun-Wu Sim (심건우)

doi:10.9708/jksci.2021.26.08.023

Comparative Analysis of Anomaly Detection Models using AE and Suggestion of Criteria for Determining Outliers

Journal of The Korea Society of Computer and Information
Abbr : JKSCI
2021, 26(8), pp.23-30
DOI : 10.9708/jksci.2021.26.08.023
Publisher : The Korean Society Of Computer And Information
Research Area : Engineering > Computer Science
Received : June 23, 2021
Accepted : July 31, 2021
Published : August 31, 2021

Gun-Ha Kang ¹, sohn jung mo ², Gun-Wu Sim ¹

¹이포즌
²(주) 이포즌

Accredited

ABSTRACT

In this study, we present a comparative analysis of major autoencoder(AE)-based anomaly detection methods for quality determination in the manufacturing process and a new anomaly discrimination criterion. Due to the characteristics of manufacturing site, anomalous instances are few and their types greatly vary. These properties degrade the performance of an AI-based anomaly detection model using the dataset for both normal and anomalous cases, and incur a lot of time and costs in obtaining additional data for performance improvement. To solve this problem, the studies on AE-based models such as AE and VAE are underway, which perform anomaly detection using only normal data. In this work, based on Convolutional AE, VAE, and Dilated VAE models, statistics on residual images, MSE, and information entropy were selected as outlier discriminant criteria to compare and analyze the performance of each model. In particular, the range value applied to the Convolutional AE model showed the best performance with AUC PRC 0.9570, F1 Score 0.8812 and AUC ROC 0.9548, accuracy 87.60%. This shows a performance improvement of an accuracy about 20%P(Percentage Point) compared to MSE, which was frequently used as a standard for determining outliers, and confirmed that model performance can be improved according to the criteria for determining outliers.

KEYWORDS

Deep Learning, Manufacture Process, Anomaly Detection, Autoencoder

Citation status

* References for papers published after 2023 are currently being built.

[report] Yeong-Tae Baek / 2018 / PCB Defect Inspection using Deep Learning / KSCI : 325~326

[thesis] Chul-jin, Park / 2016 / A study on the quality information management for preventing human errors at quality of ship / Ulsan University Graduate School of Automotive Vessel Technology

[report] Ho Young, Lee / 2019 / Changes in labour demand and acceptability due to the Fourth Industrial Revoluation / Korea Information Society Development

[journal] E. S. Lee / 2020 / Trends in AI Technology for Smart Manufacturing in the Future / Electronics and telecommunications trends 35(1) : 60~70

[report] Raghavendra Chalapathy / 2019 / Deep Learning for Anomaly Detection: A Survey / CoRR

[journal] Alberto Tellaeche Iglesias / 2021 / On Combining Convolutional Autoencoders and Support Vector Machines for Fault Detection in Industrial Textures / Sensors / MDPI AG 21(10) : 3339~

[journal] Jungsuk Kim / 2021 / Printed Circuit Board Defect Detection Using Deep Learning via A Skip-Connected Convolutional Autoencoder / Sensors / MDPI AG 21(15) : 4968~

[report] Jinwon An / 2015 / Variational Autoencoder based AnomalyDetection using Reconstruction Probability / SNU Data MiningCenter : 1~18

[confproc] Jonathan Masci / 2011 / Stacked convolutional auto-encoders for hierarchical feature extraction / International Conference on Artificial Neural Networks, ICANN : 52~59

[confproc] Omid E. David / 2016 / DeepPainter: Painter Classification Using Deep Convolutional Autoencoders / ICANN 9887 : 20~28

[other] Diederik P Kingma / 2013 / Auto-encoding variational bayes. ICLR / arXiv preprintarXiv:1312.6114

[web] / Taeu Kim / https://taeu.github.io/paper/deeplearning-paper-vae/

[journal] Shengyuan Piao / 2019 / Accuracy improvement of unet based on dilated convolution / Journal of Physics: Conference Series / IOP Publishing 1345(5)

[other] VESAL.Sulaiman / 2019 / A 2D dilated residual U-Net for multi-organ segmentation in thoracic CT / arXiv preprint arXiv:1905.07710

[web] Ravirajsinh Dabhi / Casting product image data for quality inspection / Kaggle / https://www.kaggle.com/ravirajsinh45/reallife-industrial-dataset-of-casting-product

[web] The Nature of Statistics Bigpicture / Statistics Basics Lecture 1. Mean, Deviation, Variance, Standard Deviations / https://hsm-edu.tistory.com/1182

[web] / Garlic / https://m.blog.naver.com/yk60park/222100758577

[web] tyWizar / CityWizard / https://m.blog.naver.com/PostView.naver?blogId=y4769&logNo=220505513170&proxyReferer=https:%2F%2Fwww.google.com%2F

[web] / Entropy(Information Theory) / Wikipedia / https://en.wikipedia.org/wiki/Entropy_(information_theory)

[confproc] Robert Muller / 2021 / Claudia Linnhoff-Popien, Acoustic Anomaly Detection for Machine Sounds based on ImageTransfer Learning / CVPR

[web] Kim, Chungyun / Evaluation Method of Classification Model / https://velog.io/@skyepodium/%EB%B6%84%EB%A5%98-%EB%AA%A8%EB%8D%B8-%ED%8F%89%EA%B0%80-%EB%B0%A9%EB%B2%95

[journal] 박선일 / 2016 / Application of Receiver Operating Characteristic (ROC) Curve for Evaluation of Diagnostic Test Performance / 한국임상수의학회지 / 한국임상수의학회 33(2) : 97~101

[web] / Jakub Czakon / https://neptune.ai/blog/f1-score-accuracy-roc-aucpr-auc

KJCKorea
Journal Central

Journal of The Korea Society of Computer and Information 2023 KCI Impact Factor : 0.65

Comparative Analysis of Anomaly Detection Models using AE and Suggestion of Criteria for Determining Outliers

ABSTRACT

KEYWORDS

Citation status

* References for papers published after 2023 are currently being built.

Journal of The Korea Society of Computer and Information 2023 KCI Impact Factor : 0.65

Comparative Analysis of Anomaly Detection Models using AE and Suggestion of Criteria for Determining Outliers

ABSTRACT

KEYWORDS

Statistics

Tools

Issue List

Citation status

KCI Citation Counts (3)

REFERENCES (23) * References for papers published after 2023 are currently being built.

Search PDF

Citation

* References for papers published after 2023 are currently being built.