Title: An outlier-based analysis for behaviour and anomaly identification on IoT sensors

Authors: Felipe C. Almeida; Adilson E. Guelfi; Anderson A.A. Silva; Norisvaldo Ferraz Junior; Marvin O. Schneider; Vagner L. Gava; Sergio T. Kofuji

Addresses: Instituto de Pesquisas Tecnológicas (IPT), São Paulo, Brazil ' Universidade do Oeste Paulista, São Paulo, Brazil ' Instituto de Pesquisas Tecnológicas (IPT), São Paulo, Brazil; Universidade do Oeste Paulista, São Paulo, Brazil; Universidade de São Paulo (USP), São Paulo, Brazil; Centro Universitário SENAC, São Paulo, Brazil; Universidade Paulista (UNIP), São Paulo, Brazil ' Instituto de Pesquisas Tecnológicas (IPT), São Paulo, Brazil; Universidade de São Paulo (USP), São Paulo, Brazil ' Centro Universitário SENAC, São Paulo, Brazil ' Instituto de Pesquisas Tecnológicas (IPT), São Paulo, Brazil ' Universidade de São Paulo (USP), São Paulo, Brazil

Abstract: The pervasive nature of WSN-based IoT devices provides benefits for the industry, healthcare, and other environments. Because of that, a secure network that identifies sensor measurement errors is essential. However, the sensors are sensitive to changes according to the environment in which they are installed. Therefore, this work proposes an outlier-based analysis for behaviour and anomaly identification on IoT sensors which is twofold: first, cluster the sensors based on the variance of the devices' sensors; next, identify anomalies applying Mahalanobis distance within the clusters to identify anomalous devices. We use different datasets in the experimental process to validate our proposal. As a result, we demonstrate anomalous sensors identification without relying only on the neighbour measurements (which could lead to incorrect identification of anomalies), and by segregating the sensors' behaviour based on daytime.

Keywords: IoT sensor; statistical outliers; clustering k-means; Mahalanobis distance; MD; anomaly identification.

DOI: 10.1504/IJSNET.2022.123604

International Journal of Sensor Networks, 2022 Vol.39 No.2, pp.106 - 124

Received: 30 Mar 2021
Accepted: 14 Oct 2021
Published online: 29 Jun 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article