Beyond Linear Models: Machine Learning Insights Into The Interplay Of Digital Game Addiction, Family Belonging, And Physical Activity Motivation

Abstract<br>
Aim: The aim of this study was to examine the complex relationships between psychosocial determinants such as physical activity motivation (PAM), family connectedness (FB), and digital game addiction (DGA) in adolescents using machine learning-based models and structural equation modeling (SEM). <br>
Method: This study..... In this study, the nonlinear relationships between PAM, SWB, and FB were analyzed with advanced regression models such as LOESS, piecewise regression, and multivariate adaptive regression splines (MARS). Models were constructed for both genders, outliers were removed, data were normalized, and complex interaction patterns between variables were generated using machine learning (ML). <br>
Results: According to the findings obtained from the LOESS, piecewise linear regression and MARS methods used in the study, the relationship between PAM, FB, and DGA showed different levels of success. According to the performance metrics obtained, the MARS algorithm modeled the data in the best way (MAE=4.23, MSE=22.34, RMSE=4ç73, R2=0.85). The analysis with LOESS plots showed that FB and PAM decreased as DGA increased and PAM increased as FB increased. <br>
Conclusion: As a result, it was concluded that the MARS algorithm effectively captures the complex interactions between variables such as PAM, FB, and DGA and gender-based variations thanks to its flexible structure.