Study on Using Signal Filtering Techniques for Machine Learning-based Indoor Positioning Systems(IPS)

Abstract

Internet of Things (IoT) systems, along with Machine Learning (ML) and Artificial Intelligence (AI), performed well in present systems. For location-based IoT systems, it is vital to accurately estimate the object’s geographical position to differentiate objects in an indoor environment. In this research study, Received Signal Strength Indicators (RSSI) and ML-based solutions are proposed for indoor localization. Although the RSSI-based position techniques are much more interested in position estimation, as it does not require any additional hardware, the precision remains a significant issue because of the considerable fading effects, multipath propagation, and different parameters in the indoor environments. This research study examines ML-based Indoor Positioning Systems (IPS) using different signal filtering techniques. In this work, RSSI signals are filtered separately using three filters, Moving Average, Gaussian and Median, and the impact on position estimation is observed. To examine each filter’s performance, the error is compared in terms of statistical figures of RMSE (Root Mean Squared Error) and R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> (Coefficient of Determination). Most widely used Random Forest Regression (RFR) and Extra Tree Regressor (ETR) have been used as the Supervised ML techniques, and results are compared. According to the experimental results, the above filters can reduce the position estimation error to a maximum of 12 cm, which is negligible in many IPS applications with the ETR ML technique.