Enhancing Online Child Safety through Age Detection Using Behavioral Interaction Patterns

Abstract

This study proposes an intelligent model for accurately distinguishing between children and adults based on diverse behavioral biometrics collected from touchscreen interactions. As digital platforms become increasingly widespread, the demand for effective and privacy-preserving user classification methods has grown—particularly in the context of child online protection. Behavioral features such as touch pressure, swipe velocity, gesture angle, touch frequency, distance, and timing were extracted as participants interacted with a custom-designed mobile game. The dataset included 200 real-world participants (98 children and 102 adults), and two machine learning models were employed: a Convolutional Neural Network (CNN) and a Bagging ensemble classifier. Experimental results demonstrated that both models achieved excellent performance, with the Bagging model attaining an accuracy of 99.3% and the CNN achieving 98.82%. The superior accuracy is attributed to the rich and varied set of behavioral features, which enabled the models to capture subtle differences between age groups effectively. These findings confirm the feasibility of using touch-based interaction data for age-group classification and offer a practical, non-invasive solution for enhancing child safety in digital environments. The proposed framework can be integrated into mobile applications to provide real-time age verification, particularly on platforms offering sensitive content. Moreover, the approach safeguards user privacy by eliminating the need for personally identifiable information, cameras, or microphones.