[Paper Review] Improving Prediction of Real-Time Loneliness and Companionship Type Using Geosocial Features of Personal Smartphone Data

Congyu Wu, Amanda N. Barczyk+7 more|arXiv (Cornell University)|Oct 19, 2020

Health disparities and outcomes29 references4 citations

TL;DR

This study proposes novel geosocial features from smartphone Bluetooth and GPS data to predict real-time loneliness and companionship type in college students. By analyzing the spatiotemporal patterns and social familiarity of Bluetooth encounters and location clusters, the model significantly improves loneliness prediction (AUC = 0.74) over baseline, with stronger performance for mental health than social context outcomes.

ABSTRACT

Loneliness is a widely affecting mental health symptom and can be mediated by and co-vary with patterns of social exposure. Using momentary survey and smartphone sensing data collected from 129 Android-using college student participants over three weeks, we (1) investigate and uncover the relations between momentary loneliness experience and companionship type and (2) propose and validate novel geosocial features of smartphone-based Bluetooth and GPS data for predicting loneliness and companionship type in real time. We base our features on intuitions characterizing the quantity and spatiotemporal predictability of an individual's Bluetooth encounters and GPS location clusters to capture personal significance of social exposure scenarios conditional on their temporal distribution and geographic patterns. We examine our features' statistical correlation with momentary loneliness through regression analyses and evaluate their predictive power using a sliding window prediction procedure. Our features achieved significant performance improvement compared to baseline for predicting both momentary loneliness and companionship type, with the effect stronger for the loneliness prediction task. As such we recommend incorporation and further evaluation of our geosocial features proposed in this study in future mental health sensing and context-aware computing applications.

Motivation & Objective

• To investigate the momentary relationship between loneliness and companionship type in daily life.
• To develop novel geosocial features from smartphone Bluetooth and GPS data that capture social exposure patterns.
• To evaluate the predictive power of these features for real-time loneliness and companionship type.
• To assess whether spatiotemporal predictability and social familiarity of encounters correlate with momentary loneliness.
• To support context-aware mental health interventions through passive, real-time sensing.

Proposed method

• Collected momentary self-reports of loneliness and companionship type from 129 Android-using college students over three weeks.
• Extracted geosocial features using Bluetooth proximity data and GPS location clusters to represent social exposure scenarios.
• Calculated features including number and entropy of unique Bluetooth devices, and mean/max values of social spatial and temporal regularity.
• Used sliding window prediction to evaluate model performance on real-time loneliness and companionship type classification.
• Applied regression and machine learning models with and without geosocial features to compare predictive performance.
• Updated training data daily to simulate real-world deployment and improve model adaptability.

Experimental results

Research questions

• RQ1 How are momentary loneliness levels related to the type of companionship (solitary, close-relationship, non-close-relationship) in daily life?
• RQ2 Can geosocial features derived from Bluetooth and GPS data predict real-time loneliness more accurately than baseline models?
• RQ3 Do features capturing the spatiotemporal predictability and familiarity of social encounters improve companionship type prediction?
• RQ4 Is the predictive power of geosocial features stronger for loneliness than for companionship type classification?
• RQ5 How does the inclusion of daily-updated training data affect model performance over time?

Key findings

• Loneliness was significantly lower in moments with close-relationship companions and highest in solitary moments, though being alone and feeling lonely showed only a weak correlation.
• The number and entropy of unique Bluetooth devices detected over time were significant predictors of loneliness, with higher values linked to greater loneliness.
• Features reflecting high spatiotemporal regularity in social encounters (e.g., consistent meeting times/locations) were associated with lower loneliness.
• The geosocial feature model achieved an average AUC of 0.74 for loneliness prediction, representing a 0.2 improvement over the baseline model.
• The performance gain from geosocial features was more pronounced for loneliness prediction than for companionship type classification.
• The model's predictive power improved with daily retraining, suggesting value for adaptive, real-time mental health sensing applications.