Dynamic Touchstroke Analysis with Explainable Artificial Intelligence Tree-Based Learners

As mobile devices become integral to daily life, robust authentication methods are essential for ensuring security. Traditional methods like personal identification numbers and swipe patterns remain vulnerable to social engineering attacks. To address these risks, this study investigates behavioural biometrics, specifically touch-stroke dynamics, as a transparent and secure alternative. By leveraging unique user interaction patterns, such as touch speed and pressure, this approach provides a distinctive means of authentication. Although various machine learning techniques are available for touch-stroke analysis, the interpretability of classification decisions is vital. This paper implements explainable artificial intelligence with tree-based learners, specifically decision trees and random forests, to enhance the transparency and effectiveness of touch-stroke dynamic authentication. Performance evaluations show that random forests achieve equal error rates (EER) between 0.03% and 0.05%, and decision trees yield EERs between 0.02% and 0.07%, demonstrating a balance between security and interpretability for mobile authentication.