Hybrid approach for Anomaly Detection in Time Series Data

Ghrib, Zeineb; Jaziri, Rakia; Romdhane, Rim

doi:10.1109/ijcnn48605.2020.9207013

Cited by 19 publications

(8 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The authors in [ 26 ] used the Echo-State Network, which is a method used to train RNN where only parameters for output are learned in order to train VAEs to detect anomalies in a multivariate time series, making use of the temporal dependence in the data. A hybrid approach for anomaly detection was used in [ 27 ], where Long-Short Term Memory (LSTM)-based AEs trained on normal samples were used to extract features from both normal samples and ones containing anomalies where an SVM classifier is used for detection purposes. A squeezed Convolutional VAE (SCVAE) was modeled to detect anomalies in edge devices of IoT as described in [ 28 ], and the reconstruction probability, which is a probabilistic measure that takes into account the variability of the distribution of variables, was used to tune VAEs to detect anomalies in [ 29 ].…”

Section: Related Workmentioning

confidence: 99%

Deep Learning-Based Adaptive Compression and Anomaly Detection for Smart B5G Use Cases Operation

Sayed

Ruiz

Harb

et al. 2023

Sensors

View full text Add to dashboard Cite

The evolution towards next-generation Beyond 5G (B5G) networks will require not only innovation in transport technologies but also the adoption of smarter, more efficient operations of the use cases that are foreseen to be the high consumers of network resources in the next decades. Among different B5G use cases, the Digital Twin (DT) has been identified as a key high bandwidth-demanding use case. The creation and operation of a DT require the continuous collection of an enormous and widely distributed amount of sensor telemetry data which can overwhelm the transport layer. Therefore, the reduction in such transported telemetry data is an essential objective of smart use case operation. Moreover, deep telemetry data analysis, i.e., anomaly detection, can be executed in a hierarchical way to reduce the processing needed to perform such analysis in a centralized way. In this paper, we propose a smart management system consisting of a hierarchical architecture for telemetry sensor data analysis using deep autoencoders (AEs). The system contains AE-based methods for the adaptive compression of telemetry time series data using pools of AEs (called AAC), as well as for anomaly detection in single (called SS-AD) and multiple (called MS-AGD) sensor streams. Numerical results using experimental telemetry data show compression ratios of up to 64% with reconstruction errors of less than 1%, clearly improving upon the benchmark state-of-the-art methods. In addition, fast and accurate anomaly detection is demonstrated for both single and multiple-sensor scenarios. Finally, a great reduction in transport network capacity resources of 50% and more is obtained by smart use case operation for distributed DT scenarios.

show abstract

Section: Related Workmentioning

confidence: 99%

Deep Learning-Based Adaptive Compression and Anomaly Detection for Smart B5G Use Cases Operation

Sayed

Ruiz

Harb

et al. 2023

Sensors

View full text Add to dashboard Cite

show abstract

“…By strictly training the LSTM Autoencoder on the majority (no-adversity) time-series, outliers (time-series corresponding to adverse outcomes) will generate high reconstruction errors [33]. LSTM Autoencoders have been effectively used in fall detection [46], sensor failure prediction [42], fraud detection [19] and video surveillance [68]. LSTM Autoencoders have also shown great potential in healthcare [5], with applications in retinal eye research [57], patient subtyping [6] and healthcare fraud detection [62].…”

Section: Related Workmentioning

confidence: 99%

A Knowledge Distillation Ensemble Framework for Predicting Short- and Long-Term Hospitalization Outcomes From Electronic Health Records Data

Ibrahim

Bean

Searle

et al. 2022

IEEE J. Biomed. Health Inform.

View full text Add to dashboard Cite

The ability to perform accurate prognosis is crucial for proactive clinical decision making, informed resource management and personalised care. Existing outcome prediction models suffer from a low recall of infrequent positive outcomes. We present a highly-scalable and robust machine learning framework to automatically predict adversity represented by mortality and ICU admission and readmission from time-series of vital signs and laboratory results obtained within the first 24 hours of hospital admission. The stacked ensemble platform comprises two components: a) an unsupervised LSTM Autoencoder that learns an optimal representation of the time-series, using it to differentiate the less frequent patterns which conclude with an adverse event from the majority patterns that do not, and b) a gradient boosting model, which relies on the constructed representation to refine prediction by incorporating static features. The model is used to assess a patient's risk of adversity and provides visual justifications of its prediction. Results of three case studies show that the model outperforms existing platforms in ICU and general ward settings, achieving average Precision-Recall Areas Under the Curve (PR-AUCs) of 0.891 (95% CI: 0.878 -0.939) for mortality and 0.908 (95% CI: 0.870-0.935) in predicting ICU admission and readmission.

show abstract

“…They do so by encoding the time-series in a low dimension to capture its most representative features. Since the encoded representation is compact, it is only possible to reconstruct representative features from the input, not the specifics of the input data, including any outliers [28].These models have been effectively used in fall detection [40], sensor failure prediction [36], fraud detection [15] and video surveillance [55]. Despite their potential as solutions for healthcare problems, their use has been limited to retinal eye research [45] and fraud detection in healthcare settings [51].…”

Section: Relationship To Existing Frameworkmentioning

confidence: 99%

A Knowledge Distillation Ensemble Framework for Predicting Short and Long-term Hospitalisation Outcomes from Electronic Health Records Data

Ibrahim¹,

Bean²,

Searle³

et al. 2020

Preprint

View full text Add to dashboard Cite

The ability to perform accurate prognosis of patients is crucial for proactive clinical decision making, informed resource management and personalised care. Existing outcome prediction models suffer from a low recall of infrequent positive outcomes. We present a highly-scalable and robust machine learning framework to automatically predict adversity represented by mortality and ICU admission from time-series vital signs and laboratory results obtained within the first 24 hours of hospital admission. The stacked platform comprises two components: a) an unsupervised LSTM Autoencoder that learns an optimal representation of the time-series, using it to differentiate the less frequent patterns which conclude with an adverse event from the majority patterns that do not, and b) a gradient

show abstract

Hybrid approach for Anomaly Detection in Time Series Data

Cited by 19 publications

References 8 publications

Deep Learning-Based Adaptive Compression and Anomaly Detection for Smart B5G Use Cases Operation

Deep Learning-Based Adaptive Compression and Anomaly Detection for Smart B5G Use Cases Operation

A Knowledge Distillation Ensemble Framework for Predicting Short- and Long-Term Hospitalization Outcomes From Electronic Health Records Data

A Knowledge Distillation Ensemble Framework for Predicting Short and Long-term Hospitalisation Outcomes from Electronic Health Records Data

Contact Info

Product

Resources

About