Factors associated with post-pornography masturbation among university students: a machine-learning study.

Abstract

Background: Pornography consumption is common among university students, yet little research has examined behavioral responses occurring immediately after exposure. Such behaviors may have implications for sexual health, behavioral conditioning and psychosocial well-being. This study investigated the prevalence and correlates of post-pornography masturbation among university students in Bangladesh, and evaluated predictive performance using supervised machine-learning approaches.

Methods: A cross-sectional survey was conducted among 1179 university students who reported prior pornography use. Sociodemographic, behavioral, psychological and contextual variables were assessed using self-administered questionnaires. Multivariable logistic regression examined associations with post-pornography masturbation. Supervised machine-learning models were trained and evaluated using a training-hold-out design. Model performance was assessed using accuracy, F1-score, log loss and area under the receiver operating characteristic curve.

Results: Overall, 19.8% of participants reported masturbation following pornography use, with a higher prevalence among males than females (25.2% vs 7.1%). Pornography-use frequency showed a graded association with the outcome. Residence in university halls and cigarette smoking were independently associated with higher odds, whereas psychological distress variables were not significant after adjustment. Ensemble models outperformed distance- and margin-based classifiers. Random forest achieved the highest area under the receiver operating characteristic curve, whereas extreme gradient boosting showed strong classification performance at the default probability threshold. Pornography-use frequency emerged as the most influential predictor across models.

Conclusions: Post-pornography masturbation was associated primarily with behavioral engagement patterns and selected contextual factors within this sample. Predictive modeling indicated that a limited set of routinely collected characteristics contains meaningful associative structure, although estimates remain probabilistic rather than causal. Longitudinal and externally validated research is needed to clarify temporal relationships and improve generalizability.