Article

An Explainable Deep Learning Framework for Predicting and Interpreting Social Media Addiction Behavior 

---

Abstract

The increasing prevalence of social media addiction has become a growing concern in the digital society, as excessive use of online platforms often leads to reduced productivity, psychological distress, and loss of self-control. This study aims to classify social media users based on their level of addiction by employing a deep learning model integrated with explainable artificial intelligence techniques. Behavioral, psychological, and contextual variables were used as model inputs to identify key predictors of addictive usage patterns. The model was trained and validated using a structured dataset and achieved an overall accuracy of 93 percent, demonstrating its effectiveness and stability without overfitting. Explainability was achieved through SHAP analysis, which revealed that Productivity Loss, Frequency of Use, and Self Control were the most influential factors contributing to addiction classification. The results suggest that addiction levels are primarily shaped by behavioral and psychological patterns rather than demographic characteristics. The explainable framework provides valuable insight into how digital behaviors contribute to problematic social media use and allows for transparent interpretation of model predictions. These findings highlight the potential of combining deep learning and explainable AI to better understand, predict, and manage social media addiction, offering practical implications for the development of digital well-being interventions and responsible technology use in modern society.

Keywords: Social Media Addiction, Deep Learning, Explainable AI, Behavioral Analysis, Digital Wellbeing

How to Cite:

Setiawan, I., (2026) “An Explainable Deep Learning Framework for Predicting and Interpreting Social Media Addiction Behavior ”, Journal of Digital Society 2(1). doi: https://doi.org//JDS.147