TY - JOUR
T1 - Big Data Analytics for Event Detection in the IoT-Multicriteria Approach
AU - Granat, Janusz
AU - Batalla, Jordi Mongay
AU - Mavromoustakis, Constandinos X.
AU - Mastorakis, George
PY - 2020/5
Y1 - 2020/5
N2 - Security requirements applicable to the Internet of Things (IoT) should aim to ensure integrity, authenticity and authorization, confidentiality/privacy, nonrepudiation, and last but not least, availability. Classic data analysis algorithms are no longer valid for assuring security at all levels and a new approach to data sciences is required, which would consider the complex heterogeneous nature of the IoT, taking also into consideration, its potential to deploy cross layer for security assessment mechanisms. Furthermore, data collected from sensors should be processed and analyzed nearly in real time. The classical algorithms have two main drawbacks: 1) they deal with unidimensional data and 2) they fail to assume limited information available in the stream data processing. In this article, new solutions are discussed and presented that detect anomalies in data streams nearly in real time. Specifically, we propose: 1) an event detection method used in unidimensional data streams and relying on the event strength function, which is an extension of the typical 'True or False' decision-making scheme; 2) multiple-criteria event detection approaches based on the Dynamic Pareto Set, introducing a time-depending decision set; and 3) an anomaly detection method based on multicriteria temporal graphs, combining the dynamics of decision making and multicriteria. All the proposed algorithms are presented by means of their formal description and are illustrated with examples.
AB - Security requirements applicable to the Internet of Things (IoT) should aim to ensure integrity, authenticity and authorization, confidentiality/privacy, nonrepudiation, and last but not least, availability. Classic data analysis algorithms are no longer valid for assuring security at all levels and a new approach to data sciences is required, which would consider the complex heterogeneous nature of the IoT, taking also into consideration, its potential to deploy cross layer for security assessment mechanisms. Furthermore, data collected from sensors should be processed and analyzed nearly in real time. The classical algorithms have two main drawbacks: 1) they deal with unidimensional data and 2) they fail to assume limited information available in the stream data processing. In this article, new solutions are discussed and presented that detect anomalies in data streams nearly in real time. Specifically, we propose: 1) an event detection method used in unidimensional data streams and relying on the event strength function, which is an extension of the typical 'True or False' decision-making scheme; 2) multiple-criteria event detection approaches based on the Dynamic Pareto Set, introducing a time-depending decision set; and 3) an anomaly detection method based on multicriteria temporal graphs, combining the dynamics of decision making and multicriteria. All the proposed algorithms are presented by means of their formal description and are illustrated with examples.
KW - Anomaly detection
KW - data mining
KW - Internet of Things (IoT) security
KW - multicriteria decision making
UR - http://www.scopus.com/inward/record.url?scp=85084930519&partnerID=8YFLogxK
U2 - 10.1109/JIOT.2019.2957320
DO - 10.1109/JIOT.2019.2957320
M3 - Article
AN - SCOPUS:85084930519
SN - 2327-4662
VL - 7
SP - 4418
EP - 4430
JO - IEEE Internet of Things Journal
JF - IEEE Internet of Things Journal
IS - 5
M1 - 8920092
ER -