Ugwunna, C.O., Chukwuogo, O.E., Alabi, O.A.1, Kareem, M.K., Belonwu, T.S., Oloyede, S.O.
Corresponding email: [email protected]
A B S T R A C T
Duplication or imitation of individual keystroke rhymes is very difficult which can make it very efficient to be used for identity authentication. Over time, it is possible that the keystroke style of an individual to be learned by following keystroke information obtained when the person types text. The user’s identity can always be verified by studying the user’s keyboard input styles anytime the user uses the keyboard. The technique suggested in this study uses the keystrokes that users make while typing to verify their identities. To provide an accurate verification of whether a user is authentic or fraudulent, a model that integrates machine learning and dynamic keystroke models—Decision Tree, Random Forest, Support Vector Machine, and K-nearest Neighbors—is compared and utilized. The keystroke dynamics dataset was gathered from Kaggle and consists of 51 subjects’ keyboard dynamics data, which was collected over the course of eight sessions and six months. There are 20400 samples in all in the data. This study assessed the effectiveness of machine learning algorithms with a focus on the keystroke dynamic authentication system. Python is used for the development work, while Jupyter notebook is used as the IDE. The performance of the models for different variables is assessed using the following metrics: accuracy, error equal rate, parameter performance, threshold, training time, and testing time. According to the results, the accuracy of the Random Forest, Support Vector Machine, KNN, and Decision Tree algorithms are, respectively, 98, 97.55, 97.28, and 94.26%. Based on the comparing results, Random Forest outperforms the other models, suggesting that Random Forest can be used as the system model for Keystroke Dynamic authentication.