A Multi-path Modeling Study on the Impact of Social Media Use on Adolescents

Abstract

This study examines the complex relationships among social media consumption, sleep length, mental health, and academic performance in adolescents. This research utilizes a multi-path modelling approach that integrates Structural Equation Modelling (SEM), Backpropagation Neural Networks (BPNN), and SHAP interpretability analysis to examine 705 valid survey replies. The results demonstrate that social media screen time and sleep length significantly and similarly affect academic performance. The backpropagation neural networks (BPNN) classifier shown enhanced proficiency in detecting academically at-risk students, attaining an accuracy of 92.2% and surpassing a Random Forest standard, which is essential for early intervention. SHapley Additive exPlanations (SHAP) analysis offered clarity, validating the equitable influence of screen time and sleep while uncovering nonlinear relationships in risk buildup. Nonetheless, the BPNN regression model exhibited reduced accuracy in forecasting mental health scores, indicating that emotional well-being is influenced by intricate psychosocial aspects beyond just behavioral measurements. This study reinforces the application of interpretable machine learning for the surveillance of teenage behavior and academic risk. It underscores the necessity of incorporating larger variables such as familial and emotional circumstances in future research to develop more comprehensive predictive models.