Novel Explainable CNN-LightGBM Model for Smartphone Continuous Authentication
Damandeep Kaur, Glaucio H.S. Carvalho, Renata Dividino, Isaac Woungang, Alagan Anpalagan · IEEE Access · 2026
Smartphones have become integral to our everyday activities, from storing significant amounts of private information to serving as access points to various online services. However, one-time authentication techniques such as Personal Identification Number (PIN) and passwords are no longer sufficient to keep our data secure as they do not ensure user identity verification after the initial login. This highlights the importance of implementing continuous authentication as an essential component of mobile security. This paper proposes a novel, explainable device-level continuous authentication system for smartphones that combines a Convolutional Neural Network (CNN) with a Light Gradient Boosting Machine (LightGBM). The system continuously verifies user identity by analyzing behavioral patterns captured through built-in smartphone sensors. Using the ExtraSensory dataset, our model achieves an average accuracy of 98.7% and an average Equal Error Rate (EER) of 2.07%. This performance is competitive with or surpasses that of other state-of-the-art systems, including Adaptive Recurrent Neural Network (Adaptive-RNN), Polynomial Neural Networks and deep neural network-based approaches, under similar evaluation protocols. Additionally, to address the critical challenge of explainability in continuous authentication, the Local Interpretable Model-agnostic Explanations (LIME) Explainable Artificial Intelligence (XAI) technique has been incorporated, ensuring transparency in the model’s decision-making process through multiple instances of both legitimate and imposter classes.We additionally validate our findings using SHAP (SHapley Additive exPlanations) to ensure robustness of explanations.