Large-scale Statistical Inference in Internet of Things Scenarios
Loading...
Journal Title
Journal ISSN
Volume Title
Perustieteiden korkeakoulu |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Author
Date
2019-08-20
Department
Major/Subject
Applied and Engineering Mathematics
Mcode
SCI3016
Degree programme
Master’s Programme in Applied and Engineering Mathematics (N5TeAM)
Language
en
Pages
75 + 10
Series
Abstract
The problem we address is performing statistical inference in Internet of Things scenarios, which are typically composed by a massive number of sensor nodes. We consider a traditional distributed scenario in which sensors make local observations about monitored phenomena and supply a central node (fusion centre) with relevant statistics for making a global decision about the state of the network. The primary contribution of the proposed statistical inference framework is that it is tailored for large-scale multisensor networks by combining non-parametric local detection approach with multiple hypotheses testing procedure that controls error rates. In particular, a local detector is employed by each sensor node. The decision problem at the detector is formulated as a binary hypothesis. The detector employs bootstrapping and non-parametric two-sample Anderson-Darling test to approximate the probability function of and calculate relevant test statistics. The fusion centre employs the False Discovery Rate control procedure for simultaneously evaluating the massive number of sensors test statistics as well as for controlling error rates. For a large number of sensors, demanding local conditions (probability models that resemble each other under the null and alternative hypotheses) and several sensors observing departure from nominal conditions, the power of the proposed statistical framework is large (above 90 %). When only few events occur, it might be challenging for the fusion centre to detect them all. To understand and address this problem we analysed the simulation results, which brought insights into the performance of the proposed statistical inference approach as well as potential ways of improving it.Description
Supervisor
Ilmonen, PauliinaThesis advisor
Ilmonen, PauliinaKeywords
statistical inference, non-parametric distributed detection, multiple hypotheses testing, false discovery rate (FDR) control, Internet of Things (IoT), wireless sensor networks
