In our work, we propose a privacy framework that helps developers design IoT solutions that follow General Data Protection Regulation (GDPR) and other privacy rules from the ground up. We guide teams through a practical process that detects privacy risks early, explains how to classify sensitive data, and suggests clear countermeasures to reduce exposure. We show how developers can use this framework to build systems that collect and process data in a safe way, even when devices operate across many environments. Our results show that a structured approach gives developers more confidence and leads to safer and more transparent IoT applications.

We propose Diff-TSD, a method that creates synthetic training data from motion sensors by turning accelerometer signals into image-like representations that capture how movements evolve over time. These images feed diffusion models, which learn to generate new high-quality samples when real labeled data are scarce. Our results show that models trained with this synthetic data keep stable accuracy even with limited real inputs. This approach supports applications in health monitoring, activity analysis, and assistive technologies, where collecting real sensor data often requires high cost or extra effort.

In our study, we review existing IoT development methodologies and analyse why many of them fail when projects scale or mix devices with different capabilities. We compare their strengths and limits and highlight gaps that still slow developers. Based on this review, we outline research directions that focus on improving modelling practices, workflow integration, and quality assurance. We show that the field still needs methods that feel simple for developers yet support the complexity of modern IoT environments. Our analysis helps guide future work toward more reliable and structured ways of building IoT systems.

We use a service-oriented and microservice-based software architecture to support stress detection in uncontrolled environments. Stress is a major cause of illness, and its automatic detection during daily activities can help prevent health problems. Unlike previous works limited to controlled settings or lacking proper labeling methods, our approach enables data collection, user labeling, and machine learning analysis across three layers of the Computing Continuum (edge, gateway, and cloud). The system also provides data visualization tools and supports the use and sharing of low-cost IoT devices. 

In our work, we explore audio-based violence detection using spectrograms and deep learning, particularly CNNs, to identify signs of violence in sound recordings. We focus on lightweight and efficient models suitable for real-world scenarios with limited computational resources. Our results show that combining visual representations of audio with deep networks improves detection accuracy while keeping the system practical and scalable. This approach supports public safety by enabling faster and automated monitoring of potentially harmful audio content. 

In our study, we design an IoT architecture that distributes data processing between the edge and the cloud to monitor stress and chronic anxiety in older adults during daily activities. We collect physiological and environmental data from sensors, along with questionnaires and clinical tests, to train machine learning models that classify stress levels. Our approach allows healthcare professionals to plan personalized interventions while exploring gender differences in stress responses. This work shows how edge computing can support real-time, scalable, and privacy-aware health monitoring.