User Localization in an Indoor Environment by Combining Different Algorithms through Plurality Rule

Abstract

User localization in an indoor environment has a wide application area including production and service systems such as factories, smart homes, hospitals, nursing homes, etc. User localization based on Wi-Fi signals has been widely studied using various classification algorithms. In this type of problem, several Wi-Fi routers placed in an indoor environment provide signals with different strengths depending on the location/room of the user. Most classification algorithms successfully make the localization with a high accuracy rate. However, in the current literature, there is no widely accepted 'best' algorithm for solving this problem. This study proposes the use of the plurality rule to combine several classification algorithms and obtain a single result. Plurality voting rule is an electoral system where the candidate that polls the most vote wins the election. We apply the plurality rule to the indoor localization problem and generate the Majority algorithm. The Majority algorithm takes the 'votes' of five different classification algorithms and provides a single result through plurality rule. Results show that the mean accuracy rate of the Majority algorithm is higher than the classification algorithms it combines. In addition, we show that proving a single accuracy rate is not sufficient for declaring that an algorithm is better than the other. Classification algorithms select the training and test data randomly and different divisions result in different accuracy rates. In this study, we show that comparing the classification algorithms through confidence intervals provides more accurate information.