Inferring Human Activity Using Wearable Sensors

Abstract

This paper presents methods that use data from wearable sensors, such as those found in low-cost commodity hardware, to infer the human activity (such as reading or walking) corresponding to the sensor readings. A related task is the identification of individuals based on the same data. The classification accuracy of the methods used in this work is higher than earlier work using the same dataset. Further, a significant reduction in the number of sensor data streams produces only a very small impact on this accuracy, which is a feature of practical significance due to implications for network bandwidth and energy budgets in such systems.