REPSense: On-line sensor data reduction while preserving data diversity for mobile sensing

Pervasive smartphones that embed a variety of sensors enable us to sense and learn about the physical environment around us, even the society we live in. However, the sheer volume of data collected through participatory sensing can deeply hamper the performance of various applications (e.g., data processing time and transmitting cost). In this paper, we proposed a method to reduce the volume of sensor data while preserving the information content of the original data. Our proposed method REPresentative Sense (REPSense) borrows the idea from electoral system. Hence, after data reduction, output data (target) can represent the original data (source) as parliament members are elected to delegate their constituencies. This method can compress multi-dimensional data with arbitrary distribution. We evaluated our method using real-world datasets collected by 12 users over a period of 4 months. The results show that our method outperforms state-of-the-art by comparing baseline methods in terms of data divergence and data processing performance.