Novel Machine Learning Techniques Based on Activities for Huntington's Disease Early Recognition and Prediction using Wearable Sensor Data

Resumen

The progressive neurodegenerative disorder Huntington's disease requires both early detection and correct prediction methods to enhance patient outcomes. The combination of recent advancements in wearable sensor systems and machine learning technologies allows for continuous tracking of digital phenotyping which identifies neurological dysfunction through actual movements. The research introduces a new framework which investigates four distinct activity-based methods that include circadian movement variation, synthetic mobility index, signal entropy, and dual-task cross-correlation to achieve both sensitive early detection and reliable HD prediction. Our pipeline uses open-source IMU datasets to extract and analyze sensor features that have not been studied in HD research. The experimental results show strong links between disease stages and motor abilities which demonstrate that advanced movement analysis can serve as a digital biomarker. The technical performance and interpretability of the methods are established through comparative modeling and detailed visual presentation. The study closes with recommendations for clinical integration, further research, and the need for larger, multi-modal longitudinal datasets.