Analysis of Machine Learning Algorithms for RFID Based 2D Indoor Localization

Abstract

Nowadays, radio frequency identification (RFID) localization techniques have been widely used in indoor positioning systems (IPS) due to their low cost and ease of deployment. The main reason for the rise in RFID localization is because of inaccuracies faced by the global positioning system (GPS) in the indoor environment due to multi-path interferences of signals. The localization methodology based on received signal strength indication (RSSI) technology for indoor RFID is currently a hot topic. Because RSSI obtained is highly dependent on environments, classic algorithms like trilateration will lead to huge errors in the accuracy of the localization. This paper introduces a novel approach for RFID based indoor localization by making use of machine learning algorithms such as artificial neural network (ANN), support vector machines (SVM) and K-nearest neighbors (KNN). Hyperparameter tuning is incorporated for increasing the accuracy of the models. Experimental results show that the ANN algorithm remarkably improves the indoor localization accuracy and is also effective in tackling nonlinear changes in radio frequency signals. Moreover, the proposed model can be used in similar environments.