2026-09-19T03:49:06Zhttps://dspace.mit.edu/server/oai/requestoai:dspace.mit.edu:1721.1/1450012026-02-02T14:53:01Zcom_1721.1_7582com_1721.1_7581col_1721.1_131023
Veeramachaneni, Kalyan
Marzouk, Youssef
Alnegheimish, Sarah(Sarah Abdulaziz)
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Massachusetts Institute of Technology. Center for Computational Science and Engineering
2022-08-29T16:26:27Z
2022-08-29T16:26:27Z
2022-05
2022-06-07T16:56:45.125Z
https://hdl.handle.net/1721.1/145001
With the recent proliferation of temporal observation data comes an increasing demand for time series anomaly detection. New methods to detect anomalies using machine learning are continuously emerging. However, algorithms alone only solve one aspect of the problem – finding anomalies. Existing systems often fail to encompass an end-to-end detection process, to facilitate comparative analysis of various anomaly detection methods, or to incorporate human knowledge to refine output. This precludes current methods from being used in real-world settings by practitioners who are not machine learning experts.
In this thesis, we introduce Orion, a machine learning framework for unsupervised time series anomaly detection. The framework supports all the steps of the anomaly detection process. It includes a pipeline hub to maintain many state-of-the-art approaches for time series anomaly detection including statistical and machine learning based methods. Orion logs the entire anomaly detection journey, providing detailed documentation of the status of a signal and anomalies over time. It enables users to analyze signals, compare methods, and investigate anomalies through an interactive visualization tool, where they can annotate events by modifying existing events, creating new ones, and removing them. Using these annotations, the framework aims to leverage human knowledge to improve the performance of the pipeline. We demonstrate the effectiveness and efficiency of Orion through a series of experiments from benchmarking to AutoML on three public time series datasets: NASA, Yahoo, and Numenta. In addition, we showcase the usability of our framework through a study conducted on a real-world use case involving spacecraft experts tasked with anomaly analysis tasks. Orion’s framework, code, and datasets are open-sourced at https://github.com/sintel-dev/Orion.
S.M.
S.M.
Massachusetts Institute of Technology
In Copyright - Educational Use Permitted
Copyright MIT
http://rightsstatements.org/page/InC-EDU/1.0/
Orion – A Machine Learning Framework for Unsupervised Time Series Anomaly Detection
Thesis
application/pdf
Master
Master of Science in Computational Science and Engineering
Master of Science in Electrical Engineering and Computer Science
Publication
unknown
unknown
<Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="45bef537-2a23-4a73-976d-99eb17d777d8">
<Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843</Type>
<Title>Orion – A Machine Learning Framework for Unsupervised Time Series Anomaly Detection</Title>
<PublishedIn>
<Publication>
</Publication>
</PublishedIn>
<PublicationDate>2022-05</PublicationDate>
<Authors>
<Author>
<DisplayName>Alnegheimish, Sarah(Sarah Abdulaziz)</DisplayName>
<Affiliation>
<OrgUnit>
</OrgUnit>
</Affiliation>
</Author>
</Authors>
<Editors>
</Editors>
<Publishers>
<Publisher>
<DisplayName>Massachusetts Institute of Technology</DisplayName>
<OrgUnit />
</Publisher>
</Publishers>
<License>http://rightsstatements.org/page/InC-EDU/1.0/</License>
<Abstract>With the recent proliferation of temporal observation data comes an increasing demand for time series anomaly detection. New methods to detect anomalies using machine learning are continuously emerging. However, algorithms alone only solve one aspect of the problem – finding anomalies. Existing systems often fail to encompass an end-to-end detection process, to facilitate comparative analysis of various anomaly detection methods, or to incorporate human knowledge to refine output. This precludes current methods from being used in real-world settings by practitioners who are not machine learning experts.
In this thesis, we introduce Orion, a machine learning framework for unsupervised time series anomaly detection. The framework supports all the steps of the anomaly detection process. It includes a pipeline hub to maintain many state-of-the-art approaches for time series anomaly detection including statistical and machine learning based methods. Orion logs the entire anomaly detection journey, providing detailed documentation of the status of a signal and anomalies over time. It enables users to analyze signals, compare methods, and investigate anomalies through an interactive visualization tool, where they can annotate events by modifying existing events, creating new ones, and removing them. Using these annotations, the framework aims to leverage human knowledge to improve the performance of the pipeline. We demonstrate the effectiveness and efficiency of Orion through a series of experiments from benchmarking to AutoML on three public time series datasets: NASA, Yahoo, and Numenta. In addition, we showcase the usability of our framework through a study conducted on a real-world use case involving spacecraft experts tasked with anomaly analysis tasks. Orion’s framework, code, and datasets are open-sourced at https://github.com/sintel-dev/Orion.</Abstract>
<Access xmlns="http://purl.org/coar/access_right"
>
</Access>
</Publication>