Secure Authentication Framework based on Adaptive Sparse Features and Enhanced LSTM Networks

Abstract

In the modern world, security has become a critical need across every domain. Among the various approaches, biometric authentication offers stronger protection compared to traditional security mechanisms. Within biometric systems, physical traits such as iris, fingerprint, and palm print, along with behavioural traits like gait, voice are widely used in all areas. Passwords have historically been the standard technique for limiting access to computer systems, however this strategy has a number of fundamental problems. Keystroke dynamics is a relatively recent biometric identification technique that offers a low-profile, reasonably priced way to make the standard login and password process more difficult. Keystroke dynamics enables user verification based on individual typing behaviour like rhythm and speed. In this study, Statistical measures including mean and standard deviation are calculated from keystroke features such as latency, duration, and digraph timing. To improve reliability and identify the optimum features, Adaptive sparse feature selection is applied, which efficiently identifies the most relevant feature subset and enhances the overall performance. For the classification phase, Enhanced LSTM models are employed to classify the users based on their typing patterns. The system’s effectiveness is evaluated using standard performance measures such as false positives, false negatives, true positives, and true negatives. Keystroke dynamics supports SDG in the way of  energy saving and reduce E-waste. The Accuracy of the system is improved while using the Adaptive Sparse with the proposed Enhanced LSTM models. This system identifies genuine users and intruders with 91% of Accuracy Rate. It is useful for more security areas.