Experimental Evaluation of the Poissoness of Real Sensor Data Traffic in the Internet of Things

Abstract

This work proposes a novel experimental and mathematical framework to determine the statistical models for the Internet of Things (IoT) data traffic. Conventionally, it is assumed that the data packet generation for IoT based applications follows a Poisson process with exponentially distributed packet inter-arrival time. Based on such generalized premise, majority of the network related theoretical and practical analysis of the IoT platforms are carried out. Based on empirical data for a smart home application recorded for over 10 weeks duration using proposed IoT subsystem, in this paper we estimate the empirical statistical distribution of the IoT data traffic generated by temperature, light intensity and motion sensors. The inter-arrival between the data packets generated from different sensing modules of the IoT smart home application subsystem is determined. The Empirical Cumulative Distribution Function (ECDF) of the estimated time duration is fitted with few of the well-established classical statistical distributions using Method of Moments (MoM) and Maximum Likelihood (ML) estimation techniques. The goodness of fit is quantified using Kolmogrov-Smirnov (KS) test. The parameters of the fitted distributions are determined as a function of the physical input parameters. The results reveal source IoT traffic does not follow a Poisson process which is conventionally assumed in the literature, but rather depends on the type of application.