The rapid growth of online gaming has created both opportunities and challenges, particularly regarding the safe participation of diverse demographic groups. While prior research has predominantly focused on monetization and retention, there is limited work on predictive analytics that promotes healthy gaming habits. (Introduction) This study presents a safe gaming analytics framework that applies a gender- and age-aware machine learning approach to predict player engagement levels. (Methods) Using Extreme Gradient Boosting (XGBoost) and a dataset of 8,095 online game players from Asia as a case study, the model achieved strong predictive performance (Accuracy = 0.908, Precision = 0.910, Recall = 0.899, F1-score = 0.904). Feature importance analysis identified weekly playtime, session frequency, and average session duration as the most influential predictors of engagement. (Results) Gender- and age-based analysis revealed distinct behavioral patterns, with younger male players displaying higher playtime intensity. These findings provide actionable insights for game developers, educators, and policymakers to design and implement safe gaming strategies that balance entertainment with digital well-being. (Discussion & Conclusion) The proposed framework can be adapted to various contexts beyond the present case study, supporting responsible and inclusive online gaming environments.