A Machine-Learning-Based Analysis of the Relationships between Loneliness Metrics and Mobility Patterns for Elderly

Area of Science:

• Gerontology
• Computer Science
• Human-Computer Interaction

Background:

• Loneliness and social isolation are significant issues affecting well-being, particularly in the elderly.
• Mobility is a key factor influencing feelings of loneliness and social isolation.
• Objective measurement of loneliness through mobility patterns offers a novel approach to understanding and potentially mitigating social isolation.

Purpose of the Study:

• To develop and evaluate a machine-learning approach for classifying user loneliness levels.
• To investigate the effectiveness of indoor and outdoor mobility patterns in predicting perceived loneliness.
• To compare the performance of different loneliness scales (Lubben Scale, UCLA Scale) when used with mobility data.

Main Methods:

• Collected indoor and outdoor mobility data from volunteers in a Finnish nursing home using Pozyx and Pico minifinder sensors.
• Extracted mobility features including distance traveled, speed, and frequently visited locations.
• Utilized the XGBoost machine learning algorithm for classification across indoor, outdoor, and combined datasets.

Main Results:

• The XGBoost model achieved high classification accuracy (90%-98%) for all data types (indoor, outdoor, combined).
• Mobility patterns like distance, speed, and visited clusters were significant predictors of loneliness.
• The Lubben Scale was more effective for indoor/outdoor data, while the UCLA Scale performed better with combined data.

Conclusions:

• Machine learning analysis of indoor and outdoor mobility patterns can accurately predict loneliness levels in the elderly.
• Mobility data provides valuable insights into social isolation experienced by individuals in care settings.
• The choice of loneliness scale impacts classification performance depending on the data modality used.