Prediction of Problematic Smartphone Use: A Machine Learning Approach.

Area of Science:

• Digital Health
• Computational Psychology
• Data Science

Background:

• Smartphone addiction is a growing concern globally.
• Predicting problematic smartphone use based on individual behavior is difficult.
• Existing methods lack predictive accuracy for identifying at-risk individuals.

Purpose of the Study:

• To investigate the potential of using mobile phone log data for predicting smartphone addiction levels.
• To evaluate the effectiveness of machine learning models in identifying problematic smartphone users.
• To determine which user characteristics are most indicative of smartphone addiction.

Main Methods:

• Utilized a dataset of 29,712 respondents from the Korea Internet and Security Agency (KISA).
• Employed machine learning techniques including decision tree, random forest, and XGBoost for predictive analysis.
• Integrated personal characteristics and smartphone usage information, analyzing 27 variables.

Main Results:

• The random forest model achieved the highest prediction accuracy at 82.59%.
• The decision tree model yielded the lowest accuracy at 74.56%.
• Demographic factors like age, job, and sex showed minimal contribution to predicting addiction levels.

Conclusions:

• Mobile phone log data can be effectively utilized to predict smartphone addiction.
• Machine learning models, particularly random forest, show promise for early detection.
• Future research should focus on incorporating more detailed log data for enhanced prediction accuracy.