Predicting personality from patterns of behavior collected with smartphones

Area of Science:

• Psychological science
• Computer science
• Human-computer interaction

Background:

• Smartphones are ubiquitous, sensor-rich devices with high global adoption rates.
• These devices collect extensive user behavioral data, raising significant privacy concerns.
• Understanding the link between smartphone behavior and personality is crucial.

Purpose of the Study:

• To investigate the predictability of Big Five personality dimensions using smartphone sensor and log data.
• To explore the extent to which various behavioral patterns correlate with personality traits.

Main Methods:

• A machine-learning approach was employed to analyze behavioral data from 624 volunteers over 30 days.
• Six classes of behavioral data were collected: communication, music consumption, app usage, mobility, overall phone activity, and diurnal activity.
• Personality prediction was assessed at both broad domain and narrow facet levels.

Main Results:

• Personality traits were predicted with significant accuracy (domain level: R² = 0.37; facet level: R² = 0.40) from smartphone behavioral data.
• Specific behavioral patterns in communication, music, app usage, mobility, and activity levels were distinctively predictive of personality.
• Prediction accuracy was comparable to that derived from social media digital footprints.

Conclusions:

• Passive collection and modeling of smartphone behavioral data can reveal individuals' private personality traits.
• Findings underscore the potential benefits for research and the significant risks for privacy and psychological targeting.
• The study highlights the dual nature of smartphone data: a valuable research tool and a privacy vulnerability.