This research introduces a two-stage, self-supervised learning approach that transforms movement data from dementia patients into meaningful behavioral insights.

• Converts time-series activity data into text sequences processed by language models
• Applies PageRank-based algorithms to identify significant behavioral patterns
• Analyzes clinical correlations with established metrics like MMSE and ADAS-COG scores
• Enables personalized healthcare interventions based on AI-detected patterns

This innovation significantly advances remote healthcare monitoring for dementia patients by transforming raw sensor data into actionable clinical insights, potentially allowing earlier interventions and more personalized care approaches.

Two-Stage Representation Learning for Analyzing Movement Behavior Dynamics in People Living with Dementia