Wi-Fi Fingerprinting using SVM Classification and Regression
Rui Zhou, Yiming Huang and Zhiqiang Li
Wi-Fi signal fingerprints are often used to infer the locations of mobile devices indoors. However, due to the complexity of indoor environment, it is extremely hard to model the dependency accurately. This paper proposes a Wi-Fi fingerprinting algorithm based on Support Vector Machines (SVM), which combines SVM classification and regression to model the unknown relationship and provides improved accuracy. During sampling and training, the indoor area is partitioned to subregions and the nonlinear relationship between signal fingerprints and locations as well as subregions are established. For positioning, SVM classifiers first determine the subregion that the mobile device is in, then SVM regression estimates the exact coordinate on the basis of classification result. Experiments achieve the correct classification rate of 94.58% and average error distance of 2.58 meter, outperforming positioning methods SVM regression, Nearest Neighbor and K-Nearest Neighbors.
Keywords: Positioning,Wi-Fi fingerprinting, RSSI, SVM, classification, regression.