Optimizing the configuration of an heterogeneous architecture of sensors for activity recognition, using the extended belief rule-based inference methodology

Smart environments are heterogeneous architectures with a broad range of heterogeneous electronic devices that are with high in processing capabilities for computing, considering low power consumption. They have the ability to record information about the behavior of the people by means of their interactions with the objects within an environment. This kind of environments are providing solutions to address some of the problems associated with the growing size and ageing of the population by means of the recognition of activities that can offer monitoring activities of daily living and adapting the environment. In order to deploy low-cost smart environments and reduce the computational complexity for activity recognition, it is a key issue to know the subset of sensors which are relevant for activity recognition. By using feature selection methods to optimize the subset of initial sensors in a smart environment, this paper proposes the adaption of the extended belief rule-based inference methodology (RIMER+) to handle data binary sensors and its use as the suitable classifier for activity recognition that keeps the accuracy of results even in situations where an essential sensor fails. A case study is presented in which a smart environment dataset for activity recognition with 14 sensors is set. Two optimizations with 7 and 10 sensors are obtained with two feature selection methods in which the adaptation of RIMER+ for smart environment provides an encouraged performance against the most popular classifiers in terms of robustness.