Early Detection of Digital Addiction Using Temporal Sequence Learning Techniques

Abstract

The rapid proliferation of digital devices and online platforms has led to a significant rise in problematic usage patterns collectively termed digital addiction. Early detection of such addiction is critical for enabling timely psychological intervention and promoting healthier digital habits. This paper presents a comprehensive framework for the early detection of digital addiction using temporal sequence learning techniques. The proposed system leverages Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRU), and Transformer-based models to analyze sequential behavioral data comprising screen time logs, application usage patterns, notification interactions, and daily digital engagement metrics. Feature engineering techniques are applied to extract meaningful temporal patterns from raw usage logs. The proposed architecture is evaluated on a synthesized dataset derived from smartphone usage studies. Experimental results demonstrate that the LSTM-GRU hybrid model achieves an accuracy of 94.7%, precision of 93.8%, recall of 95.1%, and an F1-score of 94.4%, outperforming traditional machine learning baselines including Support Vector Machines (SVM) and Random Forest. The system is integrated within a web-based monitoring dashboard to provide real-time addiction risk assessments. This study contributes a novel, data-driven methodology to the growing body of research on mental health monitoring and digital well-being.