Machine learning-based predictive modeling of mobile phone addiction among Chinese college students (Preprint)

Abstract

The high rate of mobile phone addiction among college students requires the development of predictive tools to support the development of early intervention and prevention measures.

This study aims to identify the psychosocial factors of mobile phone addiction (MPA) and how their weights vary over time, using machine learning models.

The predictive model was developed using data from Sample 1, comprising 1120 college students. The model was then tested through a four-wave survey of 194 first-year college students in Sample 2, with six months between each survey wave. Support Vector Machines were used for model development and testing.

The performance of the identified model was found to be acceptable. Depression, perceived stress, stressful life events, online social support, and suicidal ideation were all included in the final model for predicting MPA. The weight of stressful life events dropped dramatically at the 12-month follow-up and remained low after 18 months. All other factors’ weights increased at the 12-month follow-up before then decreasing at the 18-month follow-up.

The results suggest that the importance of many MPA predictors differs over time and is, therefore, context-specific. Our findings may guide mental health professionals to develop efficient intervention and selective prevention programs for MPA.