Enhancing Digital Child Safety Through a Transformer-Based Framework for Abnormal Behavior Detection

Abstract

The proliferation of digital platforms has significantly increased children's exposure to online threats, necessitating advanced and intelligent behavior monitoring systems. This study proposes a Transformer-based framework for abnormal behavior detection aimed at enhancing digital child safety. contextual relationships within sequential data, enabling superior detection of complex behavioral patterns. Leveraging attention mechanisms, the model integrates spatiotemporal features from video surveillance data to identify potential threats such as cyberbullying, online grooming, and inappropriate interactions. Trained on the ShanghaiTech Campus Dataset, the proposed model achieves a training accuracy of 96.4%, with a precision of 93.5% and recall of 94.5%, The system supports real-time analysis and proactive intervention across digital communication platforms and surveillance environments. This Transformer-based approach presents a promising direction for intelligent, context-aware child safety monitoring in an increasingly digital world