Continuous Behavioral Age Verification: Estimating Underage Users from Free-Text Typing Patterns

Abstract

Keystroke dynamics has been studied for decades as a behavioral biometric, but the vast majority of this work has been directed toward user authentication. This study instead investigates an underexplored application: estimating user age from keystroke patterns in order to identify underage users, offering a lightweight, privacy-preserving complement to existing age-verification methods. Using free-text keystroke data collected from 80 volunteers aged 16 to 60, we formulate age estimation as a binary classification problem distinguishing users under 21 from adults, and evaluate three classifiers, K-Nearest Neighbor, Support Vector Machine, and Random Forest, on a comprehensive set of timing-derived features. Our best-performing model achieves a recall of 0.86, with the two strongest classifiers averaging an F1-score of 0.71, indicating that the incidental behavioral signal present in ordinary typing carries meaningful information about a user's age. These findings point to keystroke dynamics as a promising tool for addressing the growing challenge of child safety online.
Keywords: Keystroke Dynamics; Age Verification; Machine Learning; Behavioral Biometrics; Child Online Safety.