Explaining Well-Being Predictions: An Online Study of XAI for Lifestyle ModelsAbstract

Invitée : Flore Vancompernolle Vromman (UCLouvain)

In a context where artificial intelligence (AI) plays an increasingly prominent role in society, this project explores its potential to predict well-being based on individuals' behaviors and lifestyle habits, with a specific focus on explainability as a key principle of responsible AI. Following an initial data collection, we have developed a rigorous machine learning pipeline to generate well-being predictions from lifestyle and behavioral data. The second phase of our research centers on explaining these predictions to users. To achieve this, we test four explanation techniques—visual, textual, numerical, and interactive—through an online survey to evaluate their effectiveness in increasing participants' awareness of their well-being. Ultimately, this project aims to enhance users' understanding of their mental health and could contribute to developing accessible educational and preventive tools, such as the Datagotchi Santé platform.

Discutant : Léo Mignot