Keystroke dynamics for intelligent biometric authentication with machine learning

Abstract The growing number of privacy breaches that access confidential and sensitive data has increased the demand for intelligent cybersecurity solutions. Authentication plays a crucial role in safeguarding sensitive data, and a promising authentication method is analyzing the user’s unique typing patterns to ensure access control. KeyRecs, a recent keystroke dynamics dataset containing temporal, demographic, and handedness features, has been proposed but has not yet been benchmarked. This work focuses on analyzing KeyRecs with the primary objective of assessing the feasibility of distinguishing users based on their typing patterns in a zero-effort attack scenario using Machine Learning (ML) models. The study includes an Exploratory Data Analysis, effectively identifying and removing outliers and visualizing distinctive writing patterns. In addition, an evaluation of the fixed-text subset of KeyRecs was established with three algorithms employed for binary and multi-class classification: K-Nearest Neighbours (KNN), Random Forest (RF), and Light Gradient Boosting Machine (LGBM). The results show that LGBM for binary classification achieved the best performance with 80% F1-score, and lowest False Rejection Rate (FRR) and Equal Error Rate (EER) mean values. In conclusion, the KeyRecs dataset offers potential to enhance security measures and access control in digital systems, paving the way for the use of ML in intelligent biometric authentication solutions.