User authentication is a vital part of securing digital services and preventing unauthorized users from gaining access to the system. Nowadays, organizations use Multi-Factor Authentication (MFA) to provide robust protection by utilizing two or more identity procedures instead of using Single Factor Authentication (SFA) which became less secure. Keystroke dynamics is a behavioral biometric that examines a user's typing rhythm to determine the subject's legitimacy using the system. Keystroke dynamics have a minimal implementation cost and do not need special hardware in the authentication process since the gathering of typing data is reasonably straightforward and does not involve additional effort from the user. In this work, we present an efficient approach that uses the quantile transformation that transforms data distribution into uniform distribution which significantly reduces the impact of outlier and extreme values. Histogram Gradient Boosting is employed as the primary classifier for the training and testing phase. Our proposed approach is evaluated on Carnegie Mellon University (CMU) keystroke benchmark dataset which has achieved 97.96% of average accuracy and 0.014 of average equal error rate (EER) across all subjects which outperforms all the previous advances in both machine and deep learning approaches.

The safeguarding of online services and prevention of unauthorized access by hackers rely heavily on user authentication, which is considered a crucial aspect of security. Currently, multi-factor authentication is used by enterprises to enhance security by integrating multiple verification methods rather than relying on a single method of authentication, which is considered less secure. Keystroke dynamics is a behavioral characteristic used to evaluate an individual’s typing patterns to verify their legitimacy. This technique is preferred because the acquisition of such data is a simple process that does not require any additional user effort or equipment during the authentication process. This study proposes an optimized convolutional neural network that is designed to extract improved features by utilizing data synthesization and quantile transformation to maximize results. Additionally, an ensemble learning technique is used as the main algorithm for the training and testing phases. A publicly available benchmark dataset from Carnegie Mellon University (CMU) was utilized to evaluate the proposed method, achieving an average accuracy of 99.95%, an average equal error rate (EER) of 0.65%, and an average area under the curve (AUC) of 99.99%, surpassing recent advancements made on the CMU dataset.