Abstract :
[en] Pervasive computing systems, envisioned as systems that blend with the physical environment to enhance the quality of life of its users, are rapidly becoming a not so distant reality. However, many challenges must be addressed before realizing the goal of having such computing systems as part of our everyday life. One such challenge is related to the problem of how to develop in a systematic way the software that lies behind pervasive systems, operating them and allowing them to intelligently adapt both to users' changing needs and to variations in the environment. In spite of the important strides done in recent years concerning the engineering of software that places the actual, immediate needs and preferences of users in the center of attention, to the best of our knowledge no work has been devoted to the study of the engineering process for building software for pervasive systems.
In this dissertation we focus on the engineering process to build smart software services for pervasive systems. Specifically, we first introduce as our first major contribution a model for the systematic construction of software for pervasive systems, which has been derived using analytical, evidence-based, and empirical methodologies. Then, on the basis of the proposed model, we investigate two essential mechanisms that provide support for the engineering of value-added software services for smart environments, namely the learning of users' daily routines and the continuous identification of users. For the case of learning users' daily routines, we propose what is our second main contribution: a novel approach that discovers periodic-frequent routines in event data from sensors and smart devices deployed at home. For the continuous identification of users we propose what is our third major contribution: a novel approach based on behavioral biometrics which is able to recognize identities without requiring any specific gesture, action, or activity from the users. The two approaches proposed have been extensively evaluated through studies in the lab, based on synthetic data, and in the wild, showing that they can be effectively applied to different scenarios and environments.
In sum, the engineering model proposed in this dissertation is expected to serve as a basis to further the research and development efforts in key aspects that are necessary to build value-added smart software services that bring pervasive systems closer to the way they have been envisioned. Furthermore, the approaches proposed for learning users' daily routines and recognizing users' identities in smart environments are aimed at contributing to the investigation and development of the data analytics technology necessary for the smart adaptation and evolution of the software in pervasive systems to users' needs.
