Unsupervised human process discovery in smart homes

The advances in the Internet of Things (IoT) have enabled the automation of various tasks like switching on the heating at home from work, seeing who is at your front door from the couch, supporting nurses in elderly homes, or the efficient delivery of packages. By enabling the connection between the physical and digital worlds, the IoT has shown how environments can be augmented with technology to enhance their capabilities, making them more intelligent, responsive, and adaptive. This widespread adoption of embedded systems turned pervasive (or ubiquitous) computing into reality: while sensors gather real-time data about the environment, actuators are used to automate the execution of many tasks that help the users of such environments. These environments, referred to as smart environments or smart spaces, represent an emerging class of IoT-based applications and are centered on their human users. Among smart spaces, smart homes and offices are representative examples. The goal is to enhance the quality of life, improve productivity, and provide personalized services by understanding and responding to the needs and preferences of the users, realizing the paradigm known as Ambient Intelligence (AmI). The literature presents various definitions of AmI systems and a set of distinct features that characterize them: sensitivity, responsiveness, adaptivity, ubiquity, and transparency. Sensitivity pertains to the AmI system's ability to perceive and comprehend the surrounding environment and its interaction context. Responsiveness and adaptivity, closely tied to sensitivity, indicate the system's capacity to promptly react, either proactively or reactively, to changes in the context in accordance with user preferences. Collectively, sensitivity, responsiveness, and adaptivity contribute to the overarching concept of context awareness. Lastly, the terms ubiquity and transparency directly relate to the idea of pervasive computing. Smart environments process and analyze the data collected from sensors to extract meaningful information. In this context, AmI is realized by utilizing techniques such as machine learning, artificial intelligence, and human-computer interaction (HCI). The rich data automatically collected via IoT sensors in smart spaces is used to get insights about the human behavior of the user (e.g., sleep tracking) or to perform automated actions for the user (e.g., automatically opening the blinds). For instance, current applications of human behavior monitoring in smart spaces include smart thermostats (e.g., Google Nest Learning Thermostat) and ambient assisted living (e.g., elderly fall detection systems). Modeling human activities and habits is not a simple task, due to the flexible and unstructured nature of human behavior. Recently, although it is still difficult to represent them following a precise flow of tasks, approaches have been proposed that model human habits as workflows. In particular, the research community and manufacturers have shown a great interest in applying process mining (PM) to smart spaces. Process mining is a fairly recent research discipline that combines data mining techniques with techniques used in Business Process Management (BPM), such as process modeling and process analysis. Process mining aims to extract, monitor, and improve processes based on real-world data. In particular, process discovery is a process mining technique used to discover and generate the process model describing the underlying behavior shown in the event log. The mined process model can be visualized in different forms, such as Petri nets, process flowcharts, or BPMN diagrams. Visualization helps to understand the structure and dynamics of processes within the smart space. However, even though process models could be extracted from smart space data, multiple important challenges arose. This thesis presents an overview of how some of the aforementioned research challenges are handled and to what degree they are addressed by the author.