Contenu connexe Similaire à Session 5.6 towards a semantic outlier detection framework in wireless sensor networks (20) Plus de semanticsconference (20) Session 5.6 towards a semantic outlier detection framework in wireless sensor networks1. SEMANTiCS 2017, Amsterdam, The Netherlands, September 11-14, 2017
Towards a Semantic Outlier Detection
Framework in Wireless Sensor Networks
Iker Esnaola-Gonzalez, Jesús Bermúdez, Izaskun
Fernandez, Santiago Fernandez, Aitor Arnaiz
2. © IK4-TEKNIKER 2017
Agenda
• Introduction
• Role of Semantics in Outlier
Detection
• The SemOD Framework
• Temperature Sensor use case
• Results
• Conclusions
3. © IK4-TEKNIKER 2017
Introduction
• Current datasets suffer from:
• Noisy data
• Missing data
• Outliers
• …
• Consequences:
• Complicate knowledge extraction
• Low quality mining results
• Inaccurate conclusions
• Solution: Preprocessing techniques
4. © IK4-TEKNIKER 2017
Introduction
• Outlier Detection:
• Spotting data that stand out
among other and do not have
the expected behaviour.
• What to do with them?
• Isolate and act on them (Fraud
Detection)
• Filter them out (Data Analytics)
• …
5. © IK4-TEKNIKER 2017
Role of Semantics in Outlier Detection
• Underlying semantics of data can
be exploited to detect outliers
• Is a 44ºC measurement an outlier?
It depends on the context:
• Location
• Time
• Season
• …
6. © IK4-TEKNIKER 2017
The SemOD Framework
• The Semantic Outlier Detection
(SemOD) Framework for WSNs
• Assists the data scientist in:
• Outlier Detection
• Outlier Classification
7. © IK4-TEKNIKER 2017
The SemOD Framework
• 3 main components
• The EEPSA Ontology
• The SemOD Method
• The SemOD Query
8. © IK4-TEKNIKER 2017
The SemOD Framework
• Use case: 3 Temperature sensors
located in IK4-Tekniker building
(Eibar, Spain)
• EEPSA Ontology for Semantic
Annotation
9. © IK4-TEKNIKER 2017
The SemOD Framework
• Infers sensors vulnerabilities
• For each vulnerability, a
SemOD Method is proposed
• SemOD Method: guide to
identify outliers caused by that
vulnerability
10. © IK4-TEKNIKER 2017
• 1st step: Sensor’s sun exposure
• Determines periods when sensor
may be exposed to sun
• The EEPSA Ontology infers them
• Depends on sensor location and
orientation
Temperature Sensor use case
11. © IK4-TEKNIKER 2017
Temperature Sensor use case
• 2nd step: Sunshine constraint
• Determines if sensor receives
sunshine when enough sun
• Derived from nearby sensor’s
solar irradiance and illuminance
12. © IK4-TEKNIKER 2017
Temperature Sensor use case
• 3rd step: SemOD Query generation
• Fills SemOD Query pattern with
information of previous steps
• Classifies measurements as
outliers caused due to sensor’s
sun exposure
CONSTRUCT {?obs1 rdf:type
eepsa:OutlierCausedBySolarRadiation }
FROM <myGRAPH >
WHERE {
?sensor1 sosa:observedProperty
m3-lite:Temperature .
?sensor2 sosa:observedProperty
m3-lite:Illuminance ;
eepsa:hasUnitOfMeasure m3-lite:Lux .
?obs1 sosa:isObservedBy ?sensor1 ;
eepsa:obsTime ?time ;
eepsa:obsDate ?date .
?obs2 sosa:isObservedBy ?sensor2 ;
eepsa:obsTime ?time ;
eepsa:obsDate ?date ;
sosa:hasSimpleResult ?illu
…
14. © IK4-TEKNIKER 2017
Conclusions
• SemOD Framework: Assistance in
Outlier Detection and Classification
• Exploit underlying semantics of
data, not just values.
• Not exclusive and complementary
to other outlier detection methods
• Applicable to multiple domains