Faulty and Missing Body Sensor Data Analysis

With advance of mobile sensor and wireless communication technologies, it becomes more feasible to monitor daily lives than ever. A smart phone, capable of sensing human motion as well as tracking current location, can record motion behavior as well as motion boundary for pattern study of social life. A passive or active Radio-frequency identification (RFID) based location tracing systems provide the resourceful information of the wildlife habitats. Accelerometers and gyroscopes can help to monitor daily activities of certain patient such as exercise patterns of the obesity. Regardless of how good the sensor is, missing or noisy data can be generated at any time with any reason-caused by the nature of monitoring sensor (on-body sensor) or communication failure. In data analysis, missing or noisy data generates many problems. In the last few decades, many approaches have been proposed. However, traditional approaches of handling missing data may require very complicated method and computational power to estimate the missing data. In this paper, we propose the Canonical Correlation based k weighted Angular Similarity (CkWAS) to map the missing data with reference pattern data set. Our experimental results show that the precision of finding similar motion patterns is 92% with 5% missing data and 64% with 95% missing data.