Touching the Difference: A Deep Learning Approach to Child-Adult Detection Based on Touch Gestures

Accurately distinguishing between children and adults based on smartphone interaction behaviors is essential for enabling safe and personalized digital experiences. This study addresses the need for automated child-adult detection by leveraging advanced interaction analysis techniques. We employed two distinct neural network architectures (MLP and DL4j) to classify users as children or adults. These models were trained and evaluated using three datasets: tap data, stroke data, and a combined dataset comprising both interaction types. Using a diverse dataset of 198 participants across various ages and demographics, the models extracted a set of discriminative features from raw touch data, achieving high classification accuracy across all datasets. The results highlight the effectiveness of our models: the MLP model achieved its best performance on the stroke dataset with an AUC of 90% and an EER of 19.04%, while the DL4j model reached an AUC of 91% and an EER of 17.33% on the same dataset. This research offers valuable insights for the development of personalized applications, security systems, and accessibility tools, contributing to the creation of safer and more inclusive digital environments.