# Improving Prediction of Real-Time Loneliness and Companionship Type Using Geosocial Features of Personal Smartphone Data

**Authors:** Congyu Wu, Amanda N. Barczyk, R. Cameron Craddock, Gabriella M. Harari, Edison Thomaz, Jason D. Shumake, Christopher G. Beevers, Samuel D. Gosling, David M. Schnyer  
**arXiv:** [2010.09807](https://arxiv.org/abs/2010.09807)  
**Date:** 2020-10-19  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2010.09807v1/

## Key points

- Loneliness is a common mental health symptom affected by social exposure patterns
- Study explores relationship between momentary loneliness and companionship type using data from 129 college students over three weeks
- Geosocial features based on Bluetooth and GPS data were developed and validated to predict loneliness and companionship type in real-time
- Features consider quantity and spatiotemporal predictability of Bluetooth encounters and GPS location clusters
- Regression analyses show significant correlation between geosocial features and momentary loneliness
- Sliding window prediction procedure demonstrates predictive power of geosocial features
- Geosocial features outperform baseline methods for predicting both loneliness and companionship type, particularly for loneliness
- Suggests incorporating geosocial features into mental health sensing applications and context-aware computing systems for further evaluation
- Understanding the relationship between loneliness, companionship type, and social exposure patterns is crucial for developing interventions targeting mental health issues related to loneliness.

## Summary

Loneliness is a common mental health symptom that can be affected by patterns of social exposure. In this study, researchers sought to explore the relationship between momentary loneliness and companionship type using data from 129 college students over three weeks. They also developed and validated novel geosocial features based on Bluetooth and GPS data from participants' smartphones to predict loneliness and companionship type in real-time. These features were created by considering the quantity and spatiotemporal predictability of individuals' Bluetooth encounters and GPS location clusters. The aim was to capture the personal significance of social exposure scenarios based on their temporal distribution and geographic patterns. To assess the effectiveness of these features, regression analyses were conducted to examine their statistical correlation with momentary loneliness. Additionally, a sliding window prediction procedure was used to evaluate their predictive power. The results showed that the geosocial features achieved significant performance improvement compared to baseline methods for predicting both momentary loneliness and companionship type; this effect was particularly strong for predicting loneliness. Based on these findings, it is recommended that these geosocial features be incorporated into future mental health sensing applications and context-aware computing systems for further evaluation. This study highlights the importance of understanding the relationship between loneliness, companionship type, and patterns of social exposure in order to develop interventions targeting mental health issues related to loneliness.