Abstract
Institutional environments such as healthcare facilities, research settings, and schools increasingly rely on technology to manage safety and reduce risk. One emerging approach involves the use of wearable devices that continuously collect physiological data, such as heart rate variability and electrodermal activity, to detect early signs of escalating stress or aggression. These systems are often framed as proactive tools that enable earlier intervention and reduce harm. At the same time, they introduce new challenges related to privacy, autonomy, and the interpretation of personal data. The broader problem connecting this thesis portfolio is how institutions can responsibly integrate predictive technologies to improve safety without undermining individual rights. While wearable systems offer measurable benefits in identifying behavioral risk earlier than traditional methods, they also shift how individuals are monitored, categorized, and managed within institutional settings. This creates a tension between improving safety outcomes and maintaining ethical accountability, especially in environments where individuals have limited ability to refuse monitoring.
The technical component of this thesis focuses on the development and evaluation of wearable-based systems for predicting aggression. Using physiological signals collected from wearable biosensors, the project explores whether patterns in data such as electrodermal activity, blood volume pulse, and skin temperature can be used to anticipate behavioral escalation. The methodology involved collecting second-by-second physiological data during controlled experimental tasks and aligning these signals with labeled behavioral events. Machine learning models, particularly random forest classifiers, were used to identify patterns associated with different levels of provocation and participant responses. Results showed that while models could capture some within-participant patterns, predictive accuracy decreased significantly when applied across different individuals. This suggests that physiological responses to stress and aggression are highly individualized, limiting the effectiveness of generalized models. The project ultimately concludes that while wearable systems show promise for detecting early signs of agitation, their effectiveness likely depends on personalized modeling approaches rather than one-size-fits-all solutions.
The STS research paper examines how governance frameworks address the ethical challenges introduced by wearable-based physiological monitoring. Specifically, it asks how institutional policies in research and healthcare settings respond to technologies that not only collect data but also generate behavioral predictions. Using a case-based analysis of IRB guidelines, the Belmont Report, the Common Rule, and HIPAA, the paper argues that existing governance frameworks prioritize safety and risk reduction while treating privacy primarily as an issue of data protection and consent. Across these cases, policies emphasize informed consent, confidentiality, and the justification of risks in relation to benefits. However, they often fail to address how physiological data is interpreted and used to make predictions about behavior. Drawing on the concepts of mutual shaping and agential realism, the analysis shows that wearable technologies do not simply measure aggression but actively shape how it is defined and managed. As a result, governance frameworks that focus only on data collection and access overlook the ethical implications of predictive inference, particularly in institutional settings where individuals have limited control over how technologies are deployed.
Together, these two projects contribute to a broader understanding of the challenges associated with integrating predictive wearable technologies into institutional environments. The technical work demonstrates both the potential and the limitations of using physiological data to anticipate behavioral outcomes, while the STS analysis highlights gaps in the governance structures that are meant to regulate these systems. Taken together, they suggest that improving safety through wearable monitoring requires not only technical innovation but also more comprehensive governance approaches that address how data is interpreted and used in decision-making. Future work should focus on developing personalized predictive models while also expanding policy frameworks to better account for transparency, accountability, and the ethical use of behavioral predictions.
