INTERNATIONAL JOURNAL OF RESEARCH CULTURE SOCIETY

Abstract:    Excessive screen exposure, gaming among children and youth are increasingly associated with behavioural challenges such as aggression, attention deficit, and hyperactivity. This study develops predictive models using data from 2,500 Indian participants (ages 6–25, balanced across genders) collected through surveys, parental reports, and standardized assessments. Classical statistical models showed moderate predictive capacity, with Logistic Regression achieving Accuracy = 0.75 (AUC = 0.81). Machine learning methods provided improvements: Support Vector Machine reached 0.80 accuracy (AUC = 0.86), while Random Forest (Accuracy = 0.85, AUC = 0.90) and Gradient Boosting (Accuracy = 0.87, AUC = 0.91) performed best among ML algorithms. Deep learning consistently outperformed these, with Feed-Forward Neural Networks attaining Accuracy = 0.87 (AUC = 0.90) and Recurrent Neural Networks achieving 0.88 accuracy (AUC = 0.91). The RNN with Attention mechanism delivered the strongest results (Accuracy = 0.92, AUC = 0.95, F1 = 0.91), highlighting its ability to capture temporal and sequential behavioural patterns. The findings confirm that deep learning models, particularly attention-based architectures, provide robust predictive capabilities for early identification of behavioural risks. These insights can guide parents, educators, and policymakers in designing proactive and gender-inclusive intervention strategies.