TY  - JOUR
T1  - Indoor Wireless LAN Fingerprints Parameterisation and
Classification in Academic Environments
AU - Ekpenyong, Moses E. 
JO  - Research Journal of Applied Sciences
VL  - 13
IS  - 9
SP  - 499
EP  - 512
PY  - 2018
DA  - 2001/08/19
SN  - 1815-932x
DO  - rjasci.2018.499.512
UR  - https://makhillpublications.co/view-article.php?doi=rjasci.2018.499.512
KW  - Fingerprint
KW  -indoor localisation
KW  -pattern classification
KW  -principal component analysis
KW  -received
signal strength
KW  -machine learning
AB  - In this study, we explore Machine Learning (ML) techniques to indoor Wireless Local Area Network
(WLAN) Fingerprints (FPs) parameterisation and classification in academic environments. First, relevant indoor
location (received signal strength indication and site specific) features were abstracted from the proposed area
of study (University of Uyo, Nigeria) in a previous research to serve as fingerprints to the current research.
Second, an unsupervised principal component analysis methodology was employed to produce Principal
Component Dominant Features (PCDFs) for the first three principal components (components with eigenvalues
of at least unity). These components revealed the degree of variances exhibited by the selected FPs. Third,
using three ML classifiers (Support Vector Machine: SVM, k-Nearest Neighbour: k-NN, decision tree and
Adaptive Neuro-Fuzzy Inference System: ANFIS) a classification of the PCDFs was performed. Results obtained
showed that decision tree and linear SVM classifiers were excellent at predicting large datasets an important
precursor to accommodating scalability in WLAN environments and areas with localisation challenges such
as difficult terrains, heavy interference and spatial or uneven distribution of wireless infrastructure as these
classifiers maintained high classification accuracies of above 90%. For small datasets, ANFIS gave good
classification accuracy when compared with other classifiers.
ER  -