Modélisation des incertitudes des données d’examen de sommeil dans le cadre des fonctions de croyance

This paper proposes a modeling approach for uncertainty in the assessment of sleep disorders, particularly sleep apnea, based on belief function theory. Traditional clinical methods—such as polysomnography and selfadministered questionnaires (Epworth scale, Pittsburgh Sleep Quality Index)—are subject to various biases: subjective responses, night-to-night variability, recording quality, and physiological factors such as heart rate variability. The proposed approach combines quantitative belief functions (e.g., triangular distributions) to represent physiological variability, and qualitative belief functions to model patient hesitation in questionnaire responses. Combination rules (Dempster-Shafer, Dubois- Prade) are used to integrate uncertainties from heterogeneous sources. This method respects the fuzzy and nondeterministic nature of biomedical data. Looking ahead, these uncertainties will be incorporated into a supervised learning algorithm for predicting mortality or cardiovascular risk, with explanatory variables ranked according to their reliability.