Mental health, personality, and cross-addictions as predictors of social media addiction: a machine learning longitudinal study

Problematic social media use (PSMU) refers to excessive, compulsive engagement with social media that impairs psychological functioning and wellbeing. Although past research has identified various correlates of PSMU, findings have been inconsistent. This study applied machine learning (ML) to predict PSMU risk over time and identify key predictors. Data were drawn from 276 adult social media users (aged 18–62) across Australia, the UK, New Zealand, and Canada. The Bergen Social Media Addiction Scale (BSMAS) was used to determine PSMU risk, applying clinical cutoffs of 19 and 24. ML models—including Random Forests, LASSO, Support Vector Machines, Logistic Regression, and Naïve Bayes—were trained using a broad set of predictors: demographic variables, personality traits, mental health indicators, motivational styles, coping strategies, and other behavioural addictions. Random Forests outperformed other models in predictive accuracy. The strongest predictors of PSMU at follow-up were baseline BSMAS scores, anxiety and COVID-related anxiety, low agreeableness, and disengagement coping (e.g., avoidance and escapism). These findings highlight the role of early symptoms, persistent anxiety, and maladaptive coping in maintaining PSMU. Future studies should incorporate additional predictors and test targeted interventions to reduce PSMU risk and promote mental wellbeing.