A data-driven approach for context definition in situation-aware wearable computing systems

A situation-aware wearable computing system is defined as a wearable device with the capability to perceive, comprehend, and project the situations occurring in the environment in order to adapt its behavior accordingly. For this kind of devices, the identification of complex situations related to the activities performed by users in various contexts is of great interest. This requires the capability to identify context states. A well-known technique for this task is the Context Space Theory (CST), which provides a multidimensional space representing contexts. Although powerful and lightweight, this technique has the drawback of requiring manual definition of such a space, a time-consuming process that involves domain experts. To address this issue, this work proposes a data-driven approach for defining context spaces in CST using kernel density estimation. This approach is compared with a state-of-theart expert-based CST technique and a fuzzy inference system for context representation, demonstrating superior performance on the Extrasensory dataset.