TY - JOUR
T1 - Machine Learning Methods for Monitoring of Quasi-Periodic Traffic in Massive IoT Networks
AU - Sørensen, René Brandborg
AU - Nielsen, Jimmy Jessen
AU - Popovski, Petar
PY - 2020/8
Y1 - 2020/8
N2 - One of the central problems in massive Internet-of-Things (IoT) deployments is the monitoring of the status of a massive number of links. The problem is aggravated by the irregularity of the traffic transmitted over the link, as the traffic intermittency can be disguised as a link failure and vice versa. In this article, we present a traffic model for IoT devices running quasiperiodic applications and we present unsupervised, parametric machine learning methods for online monitoring of the network performance of individual devices in IoT deployments with quasiperiodic reporting, such as smart metering, environmental monitoring, and agricultural monitoring. Two clustering methods are based on the Lomb-Scargle periodogram, an approach developed by astronomers for estimating the spectral density of unevenly sampled time series. We present probabilistic performance results for each of the proposed methods based on simulated data and compare the performance to a naïve network monitoring approach. The results show that the proposed methods are more reliable at detecting both hard and soft faults than the naïve-approach, especially, when the network outage is high. Furthermore, we test the methods on real-world data from a smart metering deployment. The methods, in particular the clustering method, are shown to be applicable and useful in a real-world scenario.
AB - One of the central problems in massive Internet-of-Things (IoT) deployments is the monitoring of the status of a massive number of links. The problem is aggravated by the irregularity of the traffic transmitted over the link, as the traffic intermittency can be disguised as a link failure and vice versa. In this article, we present a traffic model for IoT devices running quasiperiodic applications and we present unsupervised, parametric machine learning methods for online monitoring of the network performance of individual devices in IoT deployments with quasiperiodic reporting, such as smart metering, environmental monitoring, and agricultural monitoring. Two clustering methods are based on the Lomb-Scargle periodogram, an approach developed by astronomers for estimating the spectral density of unevenly sampled time series. We present probabilistic performance results for each of the proposed methods based on simulated data and compare the performance to a naïve network monitoring approach. The results show that the proposed methods are more reliable at detecting both hard and soft faults than the naïve-approach, especially, when the network outage is high. Furthermore, we test the methods on real-world data from a smart metering deployment. The methods, in particular the clustering method, are shown to be applicable and useful in a real-world scenario.
KW - Internet of Things (IoT)
KW - Lomb-Scargle
KW - Quality of Service (QoS)
KW - machine learning
KW - network monitoring
KW - unevenly spaced time series
UR - http://www.scopus.com/inward/record.url?scp=85089946076&partnerID=8YFLogxK
U2 - 10.1109/JIOT.2020.2983217
DO - 10.1109/JIOT.2020.2983217
M3 - Journal article
SN - 2327-4662
VL - 7
SP - 7368
EP - 7376
JO - IEEE Internet of Things Journal
JF - IEEE Internet of Things Journal
IS - 8
M1 - 9046825
ER -