Hybrid Xgboost-Lstm Framework for Adaptive Touch Behavior Biometrics: Addressing Behavioral Drift in Continuous Mobile Authentication

Touch behavior biometrics provide a secure and transparent answer to ongoing mobile authentication by analyzing behavior such as swipe patterns and tap patterns. Realworld applications, however, find it difficult to adapt to evolving environments and counter long-term behavioral drift. It has been seen that incorporating spatial touch features aids in enhancing accuracy for authentication, reducing equal error rates (EER) by 26.4% to 36.8 from using temporal features only. However, systems struggle with accuracy in different hand postures and usage scenarios. Behavioral drift, the unavoidable touch pattern evolution, also complicates continuous authentication. Experiments reveal accuracy decline over time, requiring repeated retraining. To tackle these challenges, we propose a hybrid method that combines eXtreme Gradient Boosting (XGBoost) with Long Short-Term Memory (LSTM) networks. XGBoost works well with high-dimensional data, while LSTMs excel in learning temporal patterns. Our method learns fine touch habits and responds to users' routines in real time. We also introduce a “behavior drift-aware updating mechanism,” offering uninterrupted updates against changes in user routine. This guarantees consistent authentication over the long run with reduced retraining and disconnections. Our work advances touch behavior biometrics to offer efficient, strong, continuous authentication on smartphones.