Attachment patterns and socio-emotional predictors of problematic social media use: a machine learning approach

Purpose The purpose of this study is to investigate the influence of four attachment dimensions – secure attachment, avoidance, anger and socio-emotional competence – on problematic social media use (PSMU) among university students, extending attachment theory within digital behavior research. Design/methodology/approach A quantitative design was used with 227 psychology undergraduates recruited through proportional stratified random sampling. Participants completed the Relationship Attachment Scale and the PSMU Scale, adapted according to Hernández et al. (2020) guidelines. Data were analyzed using machine learning algorithms – Linear Regression, Boosting Regression, Decision Tree, K-Nearest Neighbors and Random Forest – evaluated through cross-validation and performance metrics. Findings Secure attachment negatively predicted PSMU, while avoidance positively predicted it. Anger showed a modest positive effect, and socio-emotional competence was not significant. Among models, Boosting Regression provided the most accurate predictions, highlighting secure and avoidant attachment as the dominant predictors. Research limitations/implications The cross-sectional and self-report design limits causal inference and may involve response bias. Future research should use longitudinal designs and behavioral trace data and include additional predictors such as fear of missing out or depression. Practical implications Interventions should strengthen secure attachment and address avoidance tendencies, integrating attachment-based counseling into digital wellness programs. Social implications The findings of this study highlight the role of relational security in digital well-being, suggesting that improving attachment quality may mitigate problematic online engagement and foster healthier technology use among young adults. Originality/value This study broadens the attachment–PSMU literature by incorporating underexplored dimensions (anger and socio-emotional competence) and applying machine learning to enhance predictive modeling, offering novel insights for theory and practice in cyberpsychology.