Repository logo

Browsing by Author "Felizardo, Virginie dos Santos"

Now showing 1 - 1 of 1
  • Context-aware algorithms for Diabetes or Prediabetes prediction and diagnosis support in Ambient Assisted Living
    Publication . Felizardo, Virginie dos Santos; Santos, Nuno Manuel Garcia dos; Sousa, Miguel Castelo Branco Craveiro de
    The need to control diabetes provides an opportunity to develop new technological solutions for self-management and real predictions. These predictions can be useful in preventing unwanted events, such as hypoglycaemia. The patients with diabetes type 1 and some patients with diabetes type 2 commonly fear hypoglycaemia. The aim of this thesis is to develop a context-aware framework for hypoglycaemia prediction using sparse data, information fusion and classifiers’ consensus decision in a 24h hour time frame. With this approach, we contribute by proposing a hypoglycaemia prediction algorithm, in a self-management scenario, allowing it to be used by patients who perform their monitorization using a glucometer. The literature proposes glycaemia prediction algorithms using data from Continuous Glucose Monitoring (CGM) systems, but these approaches are not extensible to patients without these systems. Prediction algorithms based on discrete information are a challenge, so we proposed a novel context-aware framework for hypoglycaemia prediction based on data fusion and classifiers’ consensus decision. The fusion of additional context information with the conventional features can contribute to decrease the effect of inter- and intra-subject variability on prediction patterns. Also, the prediction decision based on classifiers’ consensus can contribute to the creation of suitable and generalised predictive algorithms. Integrating contextual and time-based features improves the accuracy on predicted hypoglycaemia. Using the classifiers’ consensus decision, 66% of the researched patients have over 90% of hypoglycaemia predicted (with 37.7% with 100% of hypoglycaemia predicted), without the increase of false positives (false alarms). This work shows the importance of data fusion and consensus decision to handle the patterns associated with hypoglycaemia risk and its prediction, however, further research is necessary to provide the necessary interpretability to the predictive models.
    2023-03Doctoral thesis Embargoed access Show more