Analysis of Machine Learning Algorithms for Wi-Fi-based Indoor Positioning System

Today, there is an increasing demand for low cost and accurate Indoor Positioning Systems (IPS) to cater to the needs of a growing customer base and provide better services. The main reason for the rise of IPS is the futility of Global Positioning Systems (GPS) indoors as the signals are highly attenuated by a number of physical objects. There are several methods, that compensate for the loss of GPS indoors, that achieve indoor positioning, a few of them being Wi-FI, LiFi, Bluetooth and RFID. Apart from using radio frequency and short-range radio frequency waves for indoor positioning, there are other methods to achieve the same using sensors such as inertial measurement unit (IMU), magnetometer, etc. This paper aims to explore the use of Wi-Fi for IPS as it is cost-effective, widely accessible and easy to set up. Among several Wi-Fi-based indoor positioning methods in existence, this paper introduces a novel approach towards improving the accuracy of low-cost Wi-Fi-based indoor positioning systems. This is achieved by selecting specific Wi-Fi channels that are non-overlapping and are subjected to least interference. The obtained received signal strength intensity (RSSI) is analyzed using machine learning algorithms such as K-Nearest Neighbours, Support Vector Machine, Artificial Neural Networks, and the best-suited algorithm is identified. This model can be used for a variety of applications such as asset tracking, industrial resource management, etc. The effectiveness of these applications using IPS depends on the area of the indoor locality together with the accuracy of the machine learning models. The accuracy of the tested model is approximately 4 square metres.