Applications of Anomaly Detection Using Deep Learning on Time Series Data
Authors:
- Van Quan NguyenVNVan Quan Nguyen
- Linh Van MaLMLinh Van Ma
- Jin-Young KimJKJin-Young Kim
- Kwangki KimKKKwangki Kim
- Jinsul KimJKJinsul Kim
- 1 August 2018
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 393-396
Abstract
In the modern world, time series data has become a critical part of many systems underlying various types that are recorded to reflect the status of objects according to the timeline. There are many kinds of research investigating to automate the process of analyzing time series data. Long Short-Term Memory (LSTM) network have been demonstrated to be a useful tool for learning sequence data. In this paper, we explore LSTM based approach to analyzing temporal data for abnormal detection. Stacked Long Short-Term Memory (LSTM) network is utilized as a predictor which is trained on normal data to learn the higher level temporal features, then such predictor is used to predict future values. An error-distribution estimation model is built to calculate the anomaly in the score of the observation. Anomalies are detected using a window-based method based on anomaly scores. To prove the promise applicable potential of our approach, we conducted the experiment on some domains (industry system, health monitor system, social based event detection system) come up with time series data including power consumption, ECG signal, and social data respectively.Keywords
This publication has 6 references indexed in Scilit:
- Two decades of SCADA exploitation: A brief history2017
- Time Series Anomaly Detection for Trustworthy Services in Cloud Computing SystemsIEEE Transactions on Big Data, 2017
- Large-Scale Unusual Time Series Detection2015
- Non-linear prediction with LSTM recurrent neural networks for acoustic novelty detection2015
- Anomaly detectionACM Computing Surveys, 2009
- Long Short-Term MemoryNeural Computation, 1997