RSSI-based Floor Localization Using Principal Component Analysis and Ensemble Extreme Learning Machine Technique

With the increasing requirement of the location based services, the floor localization which can determine the floor level of a mobile user in a building received much attentions. In this paper, a novel received signal strength indicator (RSSI) based method for floor localization via principal component analysis (PCA) and ensemble extreme learning machine (ELM) techniques is proposed. In the off-line phase, based on the developed RSSI fingerprint collection software in Android system, the RSSI measurement-floor level training data set was formed. Since too much access point (AP) feature dimension will lead to great redundant information which can increase the computational complexity and prolong the training time, the PCA feature extraction pre-processing is utilized for dimension reduction of the training set. Meanwhile, in order to improve the learning ability, ensemble ELM, rather single ELM, is proposed for classification learning and obtained the classification functions. In the on-line phase, the PCA is utilized to map the received RSSI measurement from higher-dimensional space to lower dimensional space. Then the classification functions are utilized for final floor level estimation. The experiment results show that the proposed RSSI-based method can offer a more accurate floor level estimation results than other existing algorithm.