Experimental machine learning for RSSI fingerprint in Indoor WiFi Localization

Conference proceedings article

Authors/Editors

Strategic Research Themes

Publication Details

Author list: Koovimol P., Pattaramalai S.

Publisher: Elsevier

Publication year: 2021

Start page: 1018

End page: 1021

Number of pages: 4

ISBN: 9780738111278

ISSN: 0928-4931

eISSN: 1873-0191

URL: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85112840720&doi=10.1109%2fECTI-CON51831.2021.9454865&partnerID=40&md5=7acc8dd0a835d9a41e72a1654f020953

Languages: English-Great Britain (EN-GB)

View on publisher site

Abstract

This research is an experimental machine learning (ML) based on MATLAB simulations using LSTM, BiLSTM, and GRU. All ML operate with three training options, Adam, RMSProp, and SDGM, to analyze Wi-Fi RSSI fingerprint data. Three wireless routers are setup in the same room under three different types of environment leading to get 6 pairs of matching fingerprint data. First, all combinations of MLs and options are simulated for comparing to find the suitable ML. Then some parameters are adjusted to increase the performance. As the result, the ML GRU with RMSProp has a maximum validation accuracy (VA) at 62.86% with minimum loss accuracy (LA) and validation loss (VL) about 0.0020 and 1.8002, respectively. Finally, the parameters adjusting increases VA about 33% and decreases LA and VL about 1.8605 and 2.0947, respectively. © 2021 IEEE.

Keywords

Wi-Fi Localization