A Group-Discrimination-Based Access Point Selection for WLAN Fingerprinting Localization

Access point (APs) selection approaches have been used in location fingerprinting systems to improve positioning accuracy and to reduce computational overhead. Although the interference between APs is unavoidable due to the overlapped channel, traditional methods treat APs individually by assuming independence among them. This paper proposes a novel group discriminant (GD)-based AP selection approach for improving location fingerprinting, in which the dependence between APs is considered. The proposed GD approach focuses on measuring the positioning capabilities of each group of APs instead of ranking APs based on their individual importance. It utilizes the risk function from support vector machines (SVMs) to estimate the GD value by maximizing the margin between reference locations. Moreover, this paper proposes a faster version, namely, recursive feature elimination (RFE-GD), to find a suboptimal solution of GD efficiently. This paper applies the proposed algorithms to realistic wireless local area networks (WLANs). Experimental results from two different test beds demonstrate that GD and RFE-GD outperform traditional AP selection schemes, reducing the mean localization error by 40.58%-41.13%. The experiments based on different fingerprinting approaches also confirm the advantages of the proposed algorithms.