Keystroke Authentication Using a Novel RTS Framework

In keystroke authentication, Recurrent Neural Networks (RNNs) are the go-to choice for information that arrives in a sequence. LSTMs and GRUs within RNNs excel at remembering these sequences. However, for data that's all over the place, RNNs struggle. Again, Multilayer Perceptrons handle this unordered data well, but miss the connections between pieces. Further, transformers are powerful techniques that focus on important details, but they're demanding and can't handle sequences well. Therefore, in this paper, a novel RTS (RNN Time Series) architecture is proposed for keystroke authentication. Keystroke authentication identifies the distinct typing patterns by analyzing the time intervals between keystrokes which enhances the identity and security of the user. Therefore, in this work, we have proposed a RTS framework consisting of channel-mixing block and time-mixing block that are capable to utilize past information for effective user authentication. The proposed RTS structure has key features of handling time complexity, long-term memory dependency and high computational performance. The developed RTS model achieves strong performance with high efficiency. To corroborate the efficacy of the proposed RTS architecture it is compared with the competative methods such as CNN, RNN, LSTM and GRU models on CMU Benchmark dataset and found better performance with reduced latency and memory utilization against all the aforementioned approaches. Further, the proposed RTS model offer superior performance, especially when capturing complex temporal dependencies, handling non-stationarity, while achieving computational efficiency. The proposed structure is evaluated based on performance metrics such as Accuracy, Precision, Recall, F1 Score and ROC.