Accurate indoor location awareness based on machine learning of environmental sensing data

We propose an environmental sensing platform called EnvLocal, that enables indoor location awareness by using machine learning. This research is inspired by the hypothesis that environmental sensors may capture various local features that can be used for location awareness in an indoor environment. Unlike other previous localization technologies that focus primarily on wireless communication, our proposed platform leverages environmental sensing in a smart building for location awareness. In this study, we developed a sensing platform consisting of a sensor toolkit with an environmental data server. We performed a comprehensive evaluation by measuring the long-term (524 days) data samples. After evaluating the learning models for ten locations distributed over five floors, the best result exhibits that a nearly 100% classification accuracy when using training data with an interval of one minute. In addition, the highest accuracies obtained with sampling intervals of 5, 10, 15 and 20 min were 95%, 93%, 90% and 89%, respectively. The evaluation results validates our hypothesis. Furthermore, seasonal sensitivity was investigated thoroughly, and our evaluation of real data covering four seasons shows the robustness and stability of our proposal.