Robust Classification of UWB NLOS/LOS Using Combined FCE and XGBoost Algorithms

Ultra-wideband (UWB) positioning systems are widely applied in localization. However, in complex environments with numerous obstacles, Non-Line-of-Sight (NLOS) propagation of UWB signals can occur, significantly affecting accuracy. Effective identification of Line-of-Sight (LOS) and NLOS signals is essential for precise localization. In this study, we propose a novel NLOS/LOS recognition method based on Extreme Gradient Boosting (XGBoost). The method begins with feature selection using the Pearson Correlation Coefficient to filter less correlated features of the UWB Channel Impulse Response (CIR) data, followed by outlier handling. Preliminary NLOS identification and classification are performed using Fuzzy Comprehensive Evaluation (FCE), with subsequent optimization of FCE weights via the Pattern Search algorithm (PSa). Final classification and recognition are then achieved through XGBoost. This approach, initially trained in one scenario, demonstrates seamless transition and strong generalization across six additional scenarios. Compared to traditional machine learning and neural networks, it offers lower system requirements. Experimental results on an open-source dataset covering seven scenarios show average accuracy, precision, recall, and F1 score of 93.2%, 94.8%, 90.3%, and 92.5%, respectively, highlighting the method’s superior performance and robustness in diverse environments.