WiFi indoor positioning algorithm based on machine learning

The time-varying Received Signal Strength (RSS) drastically reduces the correlation between signals and location information, which leads to degrade the indoor positioning accuracy in WiFi. And the kernel selection of Support Vector Regression (SVR) is limited by the Mercer theorem, it has a negative influence on the regressive result. In this paper, a new positioning algorithm based on Kernel Direct Discriminant Analysis (KDDA) and Relevance Vector Regression (RVR) is proposed to resolve these problems. The proposed algorithm employs KDDA to reconstruct the localization information contained in the RSS readings. The most discriminative localization features are then extracted while the redundant localization features and noise are discarded by KDDA. The extracted localization features are taken as inputs to RVR learning machine and the mapping between localization features and physical locations is established. The experimental results show that the proposed algorithm obtains more significant accuracy improvement than existing WKNN methods and KDDA-SVR algorithm.