Original title: Personalization of Affective Models to Enable Neuropsychiatric Digital Precision Health Interventions: A Feasibility Study
Authors: Ali Kargarandehkordi, Matti Kaisti, Peter Washington
The article investigates personalized models for emotion recognition in mobile therapies for children with autism spectrum disorder (ASD). Current digital interventions often rely on a single emotion recognition model for all, posing challenges for children with ASD. They explore personalized models—training one model per individual—to enhance emotion recognition. Testing on the Emognition dataset, they train personalized models for 10 individuals, each revealing unique facial features’ importance for emotion recognition. Comparing these personalized models to a generalized one trained on all 10 participants, they find that personalized models generally outperform the generalized model, scoring higher in F1-scores for 7 out of 10 individuals. However, they note limitations in cases where facial variation within an individual’s data is minimal. The study underscores the potential of personalized machine learning models for tailored digital therapies but acknowledges the need for variability in data for effective personalization.
Original article: https://arxiv.org/abs/2311.12812