Leveraging stacking ensemble learning for the classification of internet gaming disorder using IGDS9-SF

Internet Gaming Disorder (IGD) is a serious issue that needs to be identified at the right time to facilitate effective intervention measures. In this study, we suggest a new ensemble machine learning method for IGD detection based on cost-sensitive stacking with Random Forest and Stochastic Gradient Descent Classifier as base learners, and Logistic Regression as the meta-learner. The model was trained on a large public Internet Gaming Disorder Scale–Short Form (IGDS9-SF) dataset (n = 48,875) and extensively tested against an independent private test dataset (n = 635). Performance metrics achieved excellent results with 97.4% accuracy overall, 95.95% recall in detecting cases of IGD, 84.52% precision, and 89.87% F1-score for the IGD-Positive class. With stratified 5-fold cross-validation, the model generalized very well with a mean recall of 99.89% (±0.0025 standard deviation). Feature importance analysis revealed important IGDS9-SF items with high predictive power. The results point to the considerable potential of ensemble machine learning methods to resolving issues of class imbalance and in enabling the early detection of Internet Gaming Disorder (IGD) from psychometric data, thereby assisting in the provision of earlier and more effective clinical interventions to the vulnerable population.