Predicting Sociodemographic Attributes from Mobile Usage Patterns: Applications and Privacy Implications

Area of Science:

• Computer Science
• Sociology
• Human-Computer Interaction

Background:

• Mobile devices generate extensive digital footprints.
• These footprints can serve as proxies for user characteristics, including demographics.
• Predicting demographics from mobile usage offers benefits for personalization and market research.

Purpose of the Study:

• To assess the accuracy of predicting sociodemographic attributes (age, gender, income, education) from mobile usage metadata.
• To quantify the predictive power of different demographic attributes.
• To explore practical applications and privacy implications of demographic inference.

Main Methods:

• Utilized machine learning algorithms.
• Analyzed mobile usage data from 235 demographically diverse users.
• Examined prediction accuracy for age, gender, income, and education.

Main Results:

• Gender prediction achieved the highest accuracy (balanced accuracy = 0.862).
• Education level prediction was more challenging (balanced accuracy = 0.719).
• Age and income were classifiable above/below thresholds with acceptable accuracy.

Conclusions:

• Mobile usage data can predict certain demographic attributes with varying accuracy.
• Findings highlight potential benefits for targeted services and market research.
• Ethical considerations, including privacy and discrimination risks, are crucial.