Title: Predicting Generalized Anxiety Disorder (GAD-7) Among Gamers: A Comparative Study of Deep Learning and Traditional Machine Learning Regressors
Authors: Mosa M. M. Megdad, Samy S. Abu-Naser
Volume: 10
Issue: 1
Pages: 16-25
Publication Date: 2026/01/28
Abstract:
Ensuring the proactive detection of psychological distress is paramount for mental health researchers and practitioners, particularly within the rapidly growing gaming community. Early identification of elevated GAD-7 scores allows for timely interventions, mitigating the risk of long-term mental health complications and enhancing the overall well-being of digital entertainment users. This study evaluates and compares various machine learning algorithms to effectively and efficiently predict anxiety levels. The evaluated models include Linear Regression, Ridge, Lasso, SVR, Random Forest, Gradient Boosting, AdaBoost, K-Neighbors, Decision Tree, LGBM, XGB, and Deep Learning (MLP). Using a comprehensive dataset from the Kaggle repository, consisting of 13,464 observations and 55 features, the study identifies Ridge Regression as the superior regressor. The Ridge model achieved an MSE of 15.083, an MAE of 2.9448, and an Rē score of 32.59%. Notably, the Deep Learning model ranked fourth (MSE: 15.335, MAE: 2.963, RMSE: 3.916, Rē: 31.46%), with Linear Regression and Gradient Boosting also showing higher predictive performance. These results suggest that regularized linear models can offer highly efficient and accurate alternatives to complex neural networks in psychological scoring contexts."