NLOS Identification for Indoor Localization using Random Forest Algorithm

Abstract

Non-line-of-sight (NLOS) identification is a major challenge to indoor ranging and localization systems. Recently, many researchers have investigated this problem and a variety of NLOS identification approaches have been proposed. In this paper, we exploit features extracted from the channel impulse response (CIR) and implement the random forest (RF) machine learning algorithm to tackle the NLOS identification problem. We evaluate the RF algorithm against the popular least squares-support vector machine (LS-SVM) and other state-of-the-art classification algorithms in terms of identification performance and computational complexity. Our evaluation results show that the proposed algorithm outperforms other classification algorithms and achieves NLOS and LOS identification accuracy of 97.3% and 95% respectively. Furthermore, it has smaller computational complexity than the LS-SVM, making it best suitable for real-time implementation.