Abstract-In the last years, traffic over wireless networks has been increasing exponentially due to the impact of Internet of Things (IoT). IoT is transforming a wide range of services in different domains of urban life, such as environmental monitoring, home automation and public transportation. The so-called Smart City applications will introduce a set of stringent requirements, such as low latency and high mobility, since services must be allocated and instantiated on-demand simultaneously close to multiple devices at different locations. Efficient resource provisioning functionalities are needed to address these demanding constraints introduced by Smart City applications while minimizing resource costs and maximizing Quality of Service (QoS). In this article, the City of Things (CoT) framework is presented, which provides not only data collection and analysis functionalities but also automated resource provisioning mechanisms for future Smart City applications. CoT is deployed as a Smart City testbed in Antwerp (Belgium) that allows researchers and developers to easily setup and validate IoT experiments. A Smart City use case based on Air Quality Monitoring through the deployment of air quality sensors in moving cars has been presented showing the full applicability of the CoT framework for a flexible and scalable resource provisioning in the Smart City ecosystem.

Context-aware platforms consist of dynamic algorithms that take the context information into account to adapt the behavior of the applications. The relevant context information is modeled in a context model. Recently, a trend has emerged towards capturing the context in an ontology, which formally models the concepts within a certain domain, their relations and properties.Although much research has been done on the subject, the adoption of context-aware services in healthcare is lagging behind what could be expected. The main complaint made by users is that they had to significantly alter workflow patterns to accommodate the system. When new technology is introduced, the behavior of the users changes to adapt to it. Moreover, small differences in user requirements often occur between different environments where the application is deployed. However, it is difficult to foresee these * Corresponding author: Tel.: +32 9 331 49 38, Fax: +32 9 331 48 99 changes in workflow patterns and requirements at development time. Consequently, the context-aware applications are not tuned towards the needs of the users and they are required to change their behavior to accommodate the technology instead of the other way around.To tackle this issue, a self-learning, probabilistic, ontology-based frame- as a reason for nurse calls is used as a realistic scenario to evaluate the correctness and performance of the proposed framework. It is shown that correct results are achieved when the dataset contains at least 1,000 instances and the amount of noise is lower than 5%. The execution time and memory usage are also negligible for a realistic dataset, i.e., below 100 ms and 10 MB.

The complexity of continuous care settings has increased due to an ageing population, a dwindling number of caregivers and increasing costs. Electronic healthcare (eHealth) solutions are often introduced to deal with these issues.This technological equipment further increases the complexity of healthcare as the caregivers are responsible for integrating and configuring these solutions to their needs. Small differences in user requirements often occur between various environments where the services are deployed. It is difficult to capture these nuances at development time. Consequently, the services are not tuned towards the users' needs. This paper describes our experiences with extending an eHealth application with self-learning components such that it can automatically adjust its parameters at run-time to the users' needs and preferences. These components gather information about the usage of the application. This collected * Corresponding author: Tel.: +32 9 331 49 38, Fax: +32 9 331 48 99 information is processed by data mining techniques to learn the parameter values for the application. Each discovered parameter is associated with a probability, which expresses its reliability. Unreliable values are filtered. The remaining parameters and their reliability are integrated into the application.The eHealth application used is the ontology-based Nurse Call System (oNCS), which assesses the priority of a call based on the current context and assigns the most appropriate caregiver to a call. Decision trees and Bayesian networks are used to learn and adjust the parameters of the oNCS. For a realistic dataset of 1,050 instances, correct parameter values are discovered very efficiently as the components require at most 100 milliseconds execution time and 20 megabyte memory.