Using machine learning to identify the top predictors of adolescent’s interactive technology use for entertainment: Evidence from a longitudinal study

Abstract

Understanding the critical factors that underpin interactive technology use for entertainment is vital, which can provide accurate evidence to reduce the negative effects of excessive interactive technology use for entertainment among adolescents. Capitalizing on the machine learning approach, we aimed to provide a holistic understanding of how multiple personal and social-contextual factors predicted adolescents’ interactive technology use for entertainment across cross-sectional and longitudinal designs. By comparing seven machine learning algorithms, we found that the Random Forest and LightGBM outperformed others in model performance at two-time points. These two algorithms were used to assess the predictive capacity of 28 potential factors, indicating that gender and parental online supervision have been demonstrated the sustained correlates of adolescents’ interactive technology use for entertainment. The accessibility of home computers and internet access, along with peer influence, were significant predictors, particularly for interactive technology use for entertainment at T1. The interactive technology use for entertainment at T1 and teacher-student relationships were predictive factors specifically for interactive technology use for entertainment use at T2. This research underscores the strength of a multi-faceted approach, considering both personal and social factors, to understand adolescents’ technology use for entertainment, highlighting the positive role of supportive relationships.