Extremely randomized trees for Wi-Fi fingerprint-based indoor positioning

Wi-Fi fingerprint-based position estimation becomes one of the key component of many location based services since the existing Wi-Fi infrastructures in indoor environment can be utilized for user's position estimation in order to reduce the deployment cost. However, the low positioning accuracy is still a key challenge for indoor positioning system due to the environmental dynamics and noisy characteristics of the RF signal. This paper presents a robust indoor localization approach based on Wi-Fi fingerprints using extremely randomized trees as the location estimation algorithm. In this approach, the collected raw fingerprints data are preprocessed, and then fed to the proposed localization algorithm, given their capability to handle high dimensional and unbalanced data, to localize users. The evaluation results of the experiments conducted on the first publicly available multi-building multi-floor indoor localization database indicate that the proposed technique performs much better than the traditional systems in terms of localization accuracy and calibration effort. The proposed approach yielded the maximum localization rate of 91.44%.