Dynamic k nearest neighbours model for mobile user indoor positioning

This paper investigates a novel method for mobile user indoor positioning based on k Nearest Neighbours (kNN) algorithm, but with possibility of determining optimal number of neighbours dynamically in execution phase and for each positioning request separately. Proposed dynamic kNN (DkNN) model for indoor positioning is implemented using infrastructure of ubiquitous GSM (Global System for Mobile Communications) networks. Accuracy of proposed model is verified using field measurements, collected within the ground floor of the Technical Schools´ building, University of Belgrade. DkNN positioning model has shown accuracy improvements compared to standard kNN model with fixed k value. The maximum positioning error is reduced by 35% with regards to the standard kNN model.