Localization accuracy of classification techniques for Wi-Fi environments

We can often obtain information about user locations in indoor environments using signal strength readings from IEEE 802.11 (Wi-Fi) networks, if the user has a Wi-Fi equipped device. This can be very useful both to businesses and consumers. In this paper, we investigate several classification and prediction techniques for a popular localization method called “fingerprinting”. We study the techniques by using randomly generated data and data from real environments. It has been found that the Support Vector Machine classifier and the k-Nearest Neighbour classifier perform well in all our tests. Furthermore, all techniques under study demonstrate diminishing improvement in the probability of correct prediction as the number of measurements used for training or prediction increases.