Deep Learning for Continuous Recognition of Activities of Daily Living
Ainhoa Ruiz-Vitte, Alberto Comesaña, Blanca Larraga-García, Eduardo Rocón, Álvaro Gutiérrez · 2024
Pathological tremor, which impairs the ability to perform daily activities, presents challenges for current therapeutic strategies based on subjective clinical assessments and symptomatic management. This paper introduces an innovative system for classifying activities in patients with tremor, using models trained on time series data segmented into different window sizes. The system achieved an overall accuracy of 80%, although performance varied across tasks. This represents a significant advancement in the current state of continuous activity classification systems, highlighting the potential to enhance tremor monitoring and optimize medication dosage.