클라우드 시계열 데이터 관련 딥러닝 이상탐지 모델 성능 비교 연구
A Comparative Study on the Performance of Deep Learning Anomaly Detection Models for Cloud Time Series Data
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
클라우드 활용이 급속도로 확대되면서 복잡한 클라우드 네이티브 환경에서 다양한 장애가 발생하고 있다. 따라서 클라우드 시계열 기반 이상 탐지를 이용하여 보안 침입, 비정상적인 네트워크 활동, 서비스 장애 등 다양한 문제들을 해결하고자 하는 연구들이 진행하고자 한다. 이상 탐지는 시스템의 동작이 이전의 정상 상태와 크게 다른 특정 시간 지점을 파악하는 것이다. 클라우드 자원의 대용량 데이터를 처리하기 위해 기계학습 기술을 활용하기 시작하였으나 레이블이 지정된 데이터가 부족하기 때문에 대부분 이상탐지는 비지도 학습 모델인 딥러닝 방법으로 해결하려는 연구가 진행되고 있다. 본 연구에도 시계열 이상탐지 관련 선행 연구를 통해 확인된 우수한 성과를 보인 최신 딥러닝 모델들인 LSTM‐AE, GMM‐GRU‐VAE, GDN을 실제 클라우드 운영 환경에서 획득한 시계열 데이터들을 사용하여 학습시켜서 성능을 비교하였다. GDN 모델이 0.7488 AUPRC 와 0.9894 BestTS F1 점수로 가장 우수한 성능을 보였으며, 테스트 시간도 10분으로 가장 짧았다. LSTM-AE가 0.7378 AUPRC 와 0.9594 BestTS F1 점수로 그 뒤를 따른다. GMM-GRU-VAE 모델은 시계열 예측에서 LSTM-AE 보다는 나은 성능을 보였으나, AUPRC 스코어가 0.7229로 가장 낮아 전반적인 성능에서 뒤처졌다. 따라서 클라우드 인프라의 시계열 데이터 분석에서 GDN 모델이 가장 적합한 것으로 평가되며, 특히 빠른 처리 시간과 높은 예측 정확도를 요구하는 환경에서 유리한 선택임을 확인하였다.
Abstract
As cloud adoption rapidly expands, various issues are arising in complex cloud-native environments. Therefore, research is underway to address issues like security intrusions, abnormal network activity, and service disruptions using cloud time-series anomaly detection. An anomaly detection is defined as identifying a specific point in time where the system's behavior significantly deviates from its previous normal state. Although machine learning techniques have begun to be utilized to handle large volumes of cloud data, the lack of labeled data has led most anomaly detection research to rely on unsupervised learning models, specifically deep learning methods. This study also trained and compared the performance of state-of-the-art deep learning models with demonstrated effectiveness in previous time-series anomaly detection studies—namely, LSTM-AE, GMM-GRU-VAE, and GDN—using time-series data collected from an actual cloud operating environment. The GDN model achieved the highest performance with an AUPRC of 0.7488 and a BestTS F1 score of 0.9894, with the shortest testing time of just 10 minutes. LSTM-AE followed with an AUPRC of 0.7378 and a BestTS F1 score of 0.9594. The GMM-GRU-VAE model outperformed LSTM-AE in time-series prediction but ranked lowest in overall performance due to its AUPRC score of 0.7229. Therefore, GDN is evaluated as the most suitable model for time-series data analysis in cloud infrastructure, proving advantageous in environments requiring rapid processing time and high predictive accuracy.
참고문헌 (25)
- A. Berl, "Energy-efficient cloud computing", The Computer Journal, 53(7), pp.1045~1051 (2010)
- C. C. Aggarwal, "Outlier Analysis" (2017)
- I. Guyon, "An introduction to variable and feature selection", Journal of Machine Learning Research, 3, pp.1157~1182 (2003)
- S. Garc ́ıa, "Data Preprocessing in Data Mining" (2015), Springer International Publishing
- Lukas Ruff, "A Unifying Review of Deep and Shallow Anomaly Detection", Proceedings of the IEEE, pp.1~40 (2021)
- Machbase, "Deep Anomaly Detection in Time Series AI IoT", IEEE Internet of Things Journal, 7(7), pp.6481~6494 (2020)
- V. Chandola, "Anomaly detection : A survey", ACM Computing Surveys, 41(3), pp.1~58 (2009)
- V. M. Chandola Varun, "Comparative evaluation of anomaly detection techniques for sequence data", ICDM'08. Eighth IEEE International Conference on Data Mining, pp.743~748 (2008), IEEE
- B. Sun, "Building energy doctors : An spc and kalman filter-based method for system-level fault detection in hvac systems", IEEE Transactions on Automation Science and Engineering, 11(1), pp.215~229 (2014)
- K. Donghwoon, "A survey of deep learning-based network anomaly detection", Cluster Computing, pp.1~139 (2017)
- Kim Hyun-min, "Research on anomaly detection techniques through multidimensional time series analysis based on artificial intelligence that can be applied to multiple domains", pp.2021
- Seunghyeon Jeon, "Threshold Determination Method in Anomaly Detection using LSTM Autoencoder", Journal of KIIT, 21(4), pp.21~30 (2023)
- Ali H Mirza, "Computer network intrusion detection using sequential lstm neural networks autoencoders", 2018 26th signal processing and communications applications conference (SIU), pp.1~4 (2018), IEEE
- Jiehui Xu, "Anomaly transformer: Time series anomaly detection with association discrepancy", International Conference on Learning Representations (2022)
- Yifan Guo, "Multidimensional time series anomaly detection: A gru-based gaussian mixture variational autoencoder approach", Asian Conference on Machine Learning, pp.97~112 (2018), PMLR
- Pankaj Malhotra, "Long short term memory networks for anomaly detection in time series", 23rd European Symposium on Artificial Neural Networks, ESANN 2015 (2015)
- Ailin Deng, "Graph neural network-based anomaly detection in multivariate time series", Proceedings of the AAAI Conference on Artificial Intelligence, 35, pp.4027~4035 (2021)
- Yifan Guo, "Multidimensional time series anomaly detection: A gru-based gaussian mixture variational autoencoder approach", Asian Conference on Machine Learning, pp.97~112 (2018), PMLR
- Longyuan Li, "Anomaly detection of time series with smoothness-inducing sequential variational auto-encoder", ~ (2021)
- Dan Li, "Mad-gan:Multivariate anomaly detection for time series data with generative adversarial networks", International Conference on Artificial Neural Networks, pp.703~716 (2019), Springer
- Zijian Niu, "Lstmbased vae-gan for time-series anomaly detection", Sensors, 20(13), pp.3738~ (2020)
- Dennis Wagner, "TimeSeAD: Benchmarking Deep Multivariate Time-Series Anomaly Detection", OpenReview, pp.40~49 (2023)
- Ya Su, "Robust anomaly detection for multivariate time series through stochastic recurrent neural network", Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp.2828~2837 (2019)