Recognizing Semantic Locations from Smartphone Log with Combined Machine Learning Techniques

Smartphones, equipped with powerful processors, accelerometers, compasses and Global Positioning Systems (GPS) receivers, have favored the increase of location and context-based services over the last years. Many researchers have attempted to recognize user´s semantic location by various methods. The traditional semantic location recognition methods require partitioning the location and registering all the information to construct Wi-Fi map for fine localization, which is not practical in daily life and continuous attempt for recognizing semantic location makes smartphone battery last shorter time. Even worse, they have low accuracy when recognizing the locations located near to each other. In this paper, we propose a hybrid location recognition approach. The proposed method combines k-nearest neighbor with decision tree to recognize semantic locations. It consists of moving status detection, indoor/outdoor environment check and location recognition which consists of k-nearest neighbor (kNN) and decision tree. The proposed method can be used in indoor or dense urban environments where traditional approaches fail. Finally, the proposed method is practically developed over Android smartphones and tested in terms of performance. The experiments show the usefulness of the proposed semantic location recognition method.