The impact of variable neighbor numbers on Wi-Fi fingerprint-based indoor positioning using the KNN and WKNN algorithms

Abstract

Indoor positioning is an area where GNSS signals are either not available or very weak to provide sufficient positioning accuracy. We use smart mobile devices, which are technologically advanced today, as a solution to this issue. Despite the fact that they contain GNSS receivers and some also have dual-band chips, they currently do not have solutions for indoor spaces. As a result, we use Wi-Fi infrastructure, which is as widely used as GNSS. Although the purpose of its emergence is wireless communication, it is now one of the most popular indoor positioning applications. This study used the fingerprint approach, which is among the most successful methods of indoor positioning using this technology. We looked at two parameters related to both the positioning and calibration stages. The 2-meter point interval had the lowest mean errors when these parameters, which are the number of neighbors in KNN and WKNN algorithms and the point frequency in the calibration process, were examined. Furthermore, it has been observed that the KNN algorithm produces significant errors as the number of closest neighbors selected increases. Given the method's simplicity, we may conclude that the NN algorithm's results are quite respectable.