Uncertainty is present in every single prediction of Machine Learning (ML) models. Uncertainty Quantification (UQ) is arguably relevant, in particular for safety-critical applications. Prior research focused on the development of methods to quantify uncertainty; however, less attention has been given to how to leverage the knowledge of uncertainty in the process of model development. This work focused on applying UQ into practice, closing the gap of its utility in the ML pipeline and giving insights into how UQ is used to improve model development and its interpretability. We identified three main research questions: (1) How can UQ contribute to choosing the most suitable model for a given classification task? (2) Can UQ be used to combine different models in a principled manner? (3) Can visualization techniques improve UQ’s interpretability? These questions are answered by applying several methods to quantify uncertainty in both a simulated dataset and a real-world dataset of Human Activity Recognition (HAR). Our results showed that uncertainty quantification can increase model robustness and interpretability.

The number of available indoor location solutions has been growing, however with insufficient precision, high implementation costs or scalability limitations. As fingerprinting-based methods rely on ubiquitous information in buildings, the need for additional infrastructure is discarded. Still, the time-consuming manual process to acquire fingerprints limits their applicability in most scenarios. This paper proposes an algorithm for the automatic construction of environmental fingerprints on multi-storey buildings, leveraging the information sources available in each scenario. It relies on unlabelled crowdsourced data from users’ smartphones. With only the floor plans as input, a demand for most applications, we apply a multimodal approach that joins inertial data, local magnetic field and Wi-Fi signals to construct highly accurate fingerprints. Precise movement estimation is achieved regardless of smartphone usage through Deep Neural Networks, and the transition between floors detected from barometric data. Users’ trajectories obtained with Pedestrian Dead Reckoning techniques are partitioned into clusters with Wi-Fi measurements. Straight sections from the same cluster are then compared with subsequence Dynamic Time Warping to search for similarities. From the identified overlapping sections, a particle filter fits each trajectory into the building’s floor plans. From all successfully mapped routes, fingerprints labelled with physical locations are finally obtained. Experimental results from an office and a university building show that this solution constructs comparable fingerprints to those acquired manually, thus providing a useful tool for fingerprinting-based solutions automatic setup.

With the fast increase in the demand for location-based services and the proliferation of smartphones, the topic of indoor localization is attracting great interest. In indoor environments, users’ performed activities carry useful semantic information. These activities can then be used by indoor localization systems to confirm users’ current relative locations in a building. In this paper, we propose a deep-learning model based on a Convolutional Long Short-Term Memory (ConvLSTM) network to classify human activities within the indoor localization scenario using smartphone inertial sensor data. Results show that the proposed human activity recognition (HAR) model accurately identifies nine types of activities: not moving, walking, running, going up in an elevator, going down in an elevator, walking upstairs, walking downstairs, or going up and down a ramp. Moreover, predicted human activities were integrated within an existing indoor positioning system and evaluated in a multi-story building across several testing routes, with an average positioning error of 2.4 m. The results show that the inclusion of human activity information can reduce the overall localization error of the system and actively contribute to the better identification of floor transitions within a building. The conducted experiments demonstrated promising results and verified the effectiveness of using human activity-related information for indoor localization.

Infrastructure-free Indoor Positioning Systems (IPS) are becoming popular due to their scalability and a wide range of applications. Such systems often rely on deployed Wi-Fi networks. However, their usability may be compromised, either due to scanning restrictions from recent Android versions or the proliferation of 5G technology. This raises the need for new infrastructure-free IPS independent of Wi-Fi networks. In this paper, we propose the use of magnetic field data for IPS, through Deep Neural Networks (DNN). Firstly, a dataset of human indoor trajectories was collected with different smartphones. Afterwards, a magnetic fingerprint was constructed and relevant features were extracted to train a DNN that returns a probability map of a user’s location. Finally, two postprocessing methods were applied to obtain the most probable location regions. We asserted the performance of our solution against a test dataset, which produced a Success Rate of around 80%. We believe that these results are competitive for an IPS based on a single sensing source. Moreover, the magnetic field can be used as an additional information layer to increase the robustness and redundancy of current multi-source IPS.

Cardiac surgery patients are highly prone to severe complications post-discharge. Close follow-up through remote patient monitoring can help detect adverse outcomes earlier or prevent them, closing the gap between hospital and home care. However, equipment is limited due to economic and human resource constraints. This issue raises the need for efficient risk estimation to provide clinicians with insights into the potential benefit of remote monitoring for each patient. Standard models, such as the EuroSCORE, predict the mortality risk before the surgery. While these are used and validated in real settings, the models lack information collected during or following the surgery, determinant to predict adverse outcomes occurring further in the future. This paper proposes a Clinical Decision Support System based on Machine Learning to estimate the risk of severe complications within 90 days following cardiothoracic surgery discharge, an innovative objective underexplored in the literature. Health records from a cardiothoracic surgery department regarding 5 045 patients (60.8% male) collected throughout ten years were used to train predictive models. Clinicians' insights contributed to improving data preparation and extending traditional pipeline optimization techniques, addressing medical Artificial Intelligence requirements. Two separate test sets were used to evaluate the generalizability, one derived from a patient-grouped 70/30 split and another including all surgeries from the last available year. The achieved Area Under the Receiver Operating Characteristic curve on these test sets was 69.5% and 65.3%, respectively. Also, additional testing was implemented to simulate a real-world use case considering the weekly distribution of remote patient monitoring resources post-discharge. Compared to the random resource allocation, the selection of patients with respect to the outputs of the proposed model was proven beneficial, as it led to a higher number of high-risk patients receiving remote monitoring equipment.