A hybrid indoor localization system running ensemble machine learning

The need for localization in various fields of applications and the lack of efficiency in using GPS indoor leads to the development of Indoor Localization Systems. The recent rapid growth of mobile users and Wi-Fi infrastructure of modern buildings enables different methodologies to build high performance indoor localization system with minimum investment. This paper presents a novel model for indoor localization system on Android mobile devices with built-in application running ensemble learning method and artificial neural network. The system performance is enhanced with the implementation of background filters using built-in sensors. Notably, the proposed model is designed to gradually converge to location the longer the runtime. It eventually produces the correct rate of 95 percent for small-room localization with error radius of approximately 0.5 to 1 meter and the convergence time of 10 seconds at best. The developed model can run offline and optimized for embedded systems and Android devices based on pre-built models of Wi-Fi fingerprints.