Machine Learning-based Room Recognition

Nowadays smartphones can collect huge amounts of data from their surroundings with the help of embedded sensors. The combination of these sensor values, such as Wi-Fi Received Signal Strengths and magnetic field measurements, is assumed to be unique in some locations, which can be used to accurately predict smartphones’ indoor locations. In this work, we apply machine learning methods to derive the correlation between smartphones’ locations and the received Wi-Fi signal strength and sensor values, and we have developed an Android application that is able to distinguish between rooms. Our real-world experiment results show that the Voting ensemble predictor outperforms individual machine learning algorithms and it achieves an indoor room recognition accuracy of 94% in office-like environments. This work provides a coarsegrained indoor room recognition, which can be envisioned as a basis for accurate indoor positioning.