Integration of classification algorithms for indoor positioning system

Positioning applications become more popular with the advancement of location aware services. Global Positioning System is a successful solution for outdoors whereas it is not suitable for indoor environments due to the lack of line of sight for radio frequency signals. Therefore, various systems have been developed to solve the indoor positioning problem. Enhancing the performance of these systems is a critical issue. Several types of measurements and classification algorithms are employed to improve the positioning performance. The aim of this work is to enhance the performance of the indoor positioning system via the integration of different features (sensor measurements) and classification algorithms. For this purpose, firstly Wi-Fi Received Signal and magnetic field sensor values are combined to construct a hybrid fingerprint map. Then, the selected classifiers including decision tree, multi-layer perceptron, and Bayesian network are integrated using majority voting method. The test results demonstrate that the ensemble of sensor measurements and classifiers outperform the other individual classification algorithms in terms of classification accuracy. The proposed approach yielded the average distance error of 1.23 meter approximately.