NLoS recognition and feature importance analysis based on GAN-SSA-RF

In the field of indoor positioning, ultra-wideband (UWB) technology has attracted much attention due to its centimeter level positioning accuracy. The impact of signal propagation errors under Non-Line-of-Sight (NLoS) conditions on UWB positioning accuracy cannot be ignored. Firstly, in order to improve the recognition accuracy of UWB for NLoS signals, this paper proposes an algorithm GSR that integrates Generative Adversarial Network (GAN), Sparrow Search Algorithm (SSA), and Random Forest (RF). The experimental results show that the accuracy of the GSR algorithm in office scenarios reaches 92.17%, which is a certain improvement compared to traditional RF algorithms and significantly higher than the other five common machine learning algorithms. Next, SHAP (Shapley Additive Explanations) analysis was conducted to rank the importance of the features used for classification. The recognition accuracy using the four most important features reached 87.33%. This study not only improved the accuracy of NLoS recognition in UWB indoor positioning, but also quantified the importance of ten commonly used features for NLoS classification, providing reference and ideas for the development of UWB indoor positioning technology.