Regression Based Prediction of Roblox Game Popularity Using Extreme Gradient Boosting with Hyperparameter Optimization

The rapid growth of the digital gaming industry has increased the importance of predicting game popularity on user-generated content platforms such as Roblox, where diverse games and highly variable user engagement patterns create challenges in modeling long-term popularity trends. This study aims to develop a regression-based popularity prediction model using the Extreme Gradient Boosting (XGBoost) algorithm based on user interaction indicators, including visits, likes, dislikes, favorites, and active players. To investigate the effect of model optimization, hyperparameter tuning is performed using GridSearchCV. Model performance is evaluated using Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and the Coefficient of Determination (R²). Experimental results show that the baseline XGBoost model achieves an R² value of 80.74%, indicating strong capability in capturing non-linear popularity patterns. However, the optimized model yields a lower R² value of 77.71%, accompanied by slight increases in prediction error metrics, revealing that hyperparameter optimization does not always improve performance for highly skewed popularity data. Feature importance analysis further indicates that interaction-based attributes, particularly likes and dislikes, are the most influential predictors. These findings provide an important contribution to Informatics research by demonstrating the effectiveness of ensemble regression models for digital entertainment analytics while highlighting the need for critical evaluation of optimization strategies rather than assuming universal performance gains.