Classifier Selection for RF Based Indoor Positioning

Abstract

The selection of appropriate classifier is of great importance in improving the positioning accuracy and processing time for indoor positioning. In this work, an extensive analysis is carried out to determine the most appropriate classification algorithm to solve the indoor positioning problem. KIOS Research Center dataset is used in the experimental work. Principal Component Analysis method is employed together with Ranker method to determine the best features. In the next stage, the performances of Naive Bayes, Bayesian Network, Multilayer Perceptron, K-Nearest Neighbor and J48 Decision Tree, which are widely preferred classification algorithms for indoor positioning studies, are analyzed on four distinct mobile phones. The results of the analysis reveal that J48 Decision Tree is superior to the other classification algorithms in terms of both processing time and accuracy.