<![CDATA[Along with the rapid growth of the online gaming industry and the increasing complexity of player behavioral patterns, player engagement analysis has become a critical component in designing data-driven retention strategies. This study aims to identify the determinant factors influencing player engagement levels in online games and to develop an accurate classification prediction model. The study is motivated by the limitations of conventional engagement analysis approaches that rely solely on total playtime, which tend to be biased and do not fully represent player loyalty. A quantitative research approach was employed using a behavioral dataset consisting of 40,034 players, with a comparative evaluation of Naïve Bayes, Logistic Regression, and Random Forest algorithms. The data preprocessing stage included variable encoding, feature scaling, and data partitioning using a stratified train–test split to preserve class distribution. Hyperparameter optimization for the Random Forest model was performed using Grid Search with 5-fold cross-validation to objectively determine the optimal parameter combination. Experimental results demonstrate that the Random Forest algorithm achieved the best performance with a test accuracy of 91.30%, outperforming Naïve Bayes (83.61%) and Logistic Regression (82.64%). Feature importance analysis revealed that demographic factors and total playtime contributed relatively little to player engagement prediction, whereas weekly gaming session frequency and average session duration were identified as the most influential determinants. The findings suggest that player retention strategies are more effective when shifting from a duration-based approach to a frequency-based approach, encouraging consistent player interaction and fostering long-term engagement.]]>
