PCA based hybrid hyperplane margin clustering and regression for indoor WLAN localization

Abstract

Based on the Principal Component Analysis (PCA), a novel hybrid Support Vector Machine (SVM) Clustering and Regression (SVMCR) approach used for indoor Wireless Local Area Network (WLAN) localization is proposed in this paper. First of all, we rely on the SVM Clustering (SVMC) to conduct the classification for the sake of narrowing down the search space of fingerprints, as well as reducing the computation overhead. Second, the Received Signal Strength (RSS) is processed by using the PCA to extract the RSS features for localization. Finally, we use the Support Vector Regression (SVR) approach to characterize the relations of the RSS distributions and physical locations to achieve the accurate localization. Experimental results in a realistic indoor WLAN test-bed prove that the proposed approach not only reduces the computation and storage overhead, but also provides the high localization accuracy.