Evaluation of Machine Learning Prediction Models for Wifi based Indoor Positioning System

• Published in: 2023 International Conference on Computer, Information and Telecommunication Systems (CITS) pp. 01 - 07
• Main Authors: Estrada, Rebeca, Vera, Nelson, Aizaga, Xavier
• Format: Conference Proceeding
• Language: English
• Published: IEEE 10.07.2023
• Subjects: Coordinate measuring machines; Horst; Indoor Positioning Systems; Machine learning; Machine learning algorithms; OpenWRT; Performance evaluation; Position measurement; Predictive models; Received Signal Strength Intensity (RSSI); Wireless networks
• DOI: 10.1109/CITS58301.2023.10188689

Thanks to the availability of wireless network infrastructure, wireless service providers can determine the device's position by analyzing the signal strength received from them in indoor scenarios. In this work, we propose to evaluate four machine learning prediction models for indoor positioning with an Open source Wireless Infrastructure deployed in an office. The performance evaluation is based on RMSE and R-Squared for the estimated two-dimensional (2D) positions. The first approach estimates the distance from the devices to the routers using a RSSI-based machine learning prediction model and uses the Min-Max positioning algorithm to determine the 2D device's position. As the second approach, we use two independent RSSI-based machine learning models to predict each coordinate of device position. The third approach predicts first the X coordinate using one ML model trained with three RSSI measurements from the routers and then predicts Y using the predicted X coordinate and the RSSI values by means of another machine learning model. The last approach is similar to the latter but this time the Y prediction is made first, followed the prediction of X coordinate based on the predicted Y coordinate and RSSI values. Finally, we compared these models to an indoor positioning model based on a path loss propagation model and the Min-Max positioning algorithm.