dc.contributor.author |
Mokoena, Tshepiso
|
|
dc.contributor.author |
Lebogo, Ofentse
|
|
dc.contributor.author |
Dlaba, Asive
|
|
dc.contributor.author |
Marivate, Vukosi N
|
|
dc.date.accessioned |
2018-01-04T10:46:44Z |
|
dc.date.available |
2018-01-04T10:46:44Z |
|
dc.date.issued |
2017-09 |
|
dc.identifier.citation |
Mokoena, T. et al. 2017. Bringing sequential feature explanations to life. IEEE Africon 2017 Proceedings, 18-20 September 2017, Cape Town, South Africa |
en_US |
dc.identifier.isbn |
978-1-5386-2775-4 |
|
dc.identifier.uri |
http://ieeexplore.ieee.org/abstract/document/8095456/
|
|
dc.identifier.uri |
http://ieeexplore.ieee.org/abstract/document/8095456/?reload=true
|
|
dc.identifier.uri |
DOI: 10.1109/AFRCON.2017.8095456
|
|
dc.identifier.uri |
http://hdl.handle.net/10204/9932
|
|
dc.description |
Copyright: 2017 IEEE. Due to copyright restrictions, the attached PDF file only contains the abstract of the full text item. For access to the full text item, please consult the publisher's website. |
en_US |
dc.description.abstract |
In many real world applications, human analysts are not only interested in the detected anomalies but are also interested in the reasons behind why they were flagged as anomalous. However, existing anomaly detectors provide the analysts with no information about what caused the anomalies. A sequential feature explanation(SFE) of a detected data point is an ordered sequence of features which are presented to the analysts, one at a time until the information contained in the set of already presented features is enough for the analysts to make a decision. However, SFEs are yet to be available on data analysis platforms that allow users to clean their data by filtering out anomalies. We present Quirk, the first of its kind user interactive anomaly detection system that adds the human analyst in the loop. Firstly, it loads the data from the user and then uses a user selected anomaly detection algorithm to identify the anomalies in the dataset. Secondly, for each flagged data point of interest to the user, it generates its SFE and provides a sequential visual presentation of the dataset along the features specified in the SFE with the flagged data point distinctly highlighted. Lastly, Quirk allows the analyst to provide it with feedback by labelling whether the flagged data points are truly anomalous or not. With this new information, it then computes a predictive model that will automatically flag anomalous data for future analysis. We present all of Quirk’s functionalities and how it can be applied in real-world data analysis scenarios by presenting a use case of the system. |
en_US |
dc.language.iso |
en |
en_US |
dc.publisher |
IEEE |
en_US |
dc.relation.ispartofseries |
Worklist;19737 |
|
dc.subject |
Sequential feature explanations |
en_US |
dc.subject |
Anomaly explanations |
en_US |
dc.subject |
Outlier explanations |
en_US |
dc.title |
Bringing sequential feature explanations to life |
en_US |
dc.type |
Conference Presentation |
en_US |
dc.identifier.apacitation |
Mokoena, T., Lebogo, O., Dlaba, A., & Marivate, V. N. (2017). Bringing sequential feature explanations to life. IEEE. http://hdl.handle.net/10204/9932 |
en_ZA |
dc.identifier.chicagocitation |
Mokoena, Tshepiso, Ofentse Lebogo, Asive Dlaba, and Vukosi N Marivate. "Bringing sequential feature explanations to life." (2017): http://hdl.handle.net/10204/9932 |
en_ZA |
dc.identifier.vancouvercitation |
Mokoena T, Lebogo O, Dlaba A, Marivate VN, Bringing sequential feature explanations to life; IEEE; 2017. http://hdl.handle.net/10204/9932 . |
en_ZA |
dc.identifier.ris |
TY - Conference Presentation
AU - Mokoena, Tshepiso
AU - Lebogo, Ofentse
AU - Dlaba, Asive
AU - Marivate, Vukosi N
AB - In many real world applications, human analysts are not only interested in the detected anomalies but are also interested in the reasons behind why they were flagged as anomalous. However, existing anomaly detectors provide the analysts with no information about what caused the anomalies. A sequential feature explanation(SFE) of a detected data point is an ordered sequence of features which are presented to the analysts, one at a time until the information contained in the set of already presented features is enough for the analysts to make a decision. However, SFEs are yet to be available on data analysis platforms that allow users to clean their data by filtering out anomalies. We present Quirk, the first of its kind user interactive anomaly detection system that adds the human analyst in the loop. Firstly, it loads the data from the user and then uses a user selected anomaly detection algorithm to identify the anomalies in the dataset. Secondly, for each flagged data point of interest to the user, it generates its SFE and provides a sequential visual presentation of the dataset along the features specified in the SFE with the flagged data point distinctly highlighted. Lastly, Quirk allows the analyst to provide it with feedback by labelling whether the flagged data points are truly anomalous or not. With this new information, it then computes a predictive model that will automatically flag anomalous data for future analysis. We present all of Quirk’s functionalities and how it can be applied in real-world data analysis scenarios by presenting a use case of the system.
DA - 2017-09
DB - ResearchSpace
DP - CSIR
KW - Sequential feature explanations
KW - Anomaly explanations
KW - Outlier explanations
LK - https://researchspace.csir.co.za
PY - 2017
SM - 978-1-5386-2775-4
T1 - Bringing sequential feature explanations to life
TI - Bringing sequential feature explanations to life
UR - http://hdl.handle.net/10204/9932
ER -
|
en_ZA |