Our research addresses a critical issue in user authentication: the accuracy and reliability of keystroke dynamics-based systems. Specifically, we focus on the distinctive characteristics of individual letters and their influence on authentication accuracy, particularly in relation to the hold time feature. Utilizing the SelectFromModel feature selection technique alongside classifiers such as Random Forest, Support Vector Machine, Gradient Boosting, and Decision Tree Regressor, we conducted comprehensive experiments with real-world datasets. Our results identified letters like “W,” “S,” “U,” “T,” and “A” as particularly effective in capturing unique typing behaviors. These letters consistently showed high selection frequencies across classifiers, enhancing authentication accuracy. Although accuracy slightly declines with fewer features, the top-selected features maintain high accuracy levels, demonstrating the efficiency of our feature selection method in simplifying models without significant performance loss. Furthermore, our study highlights the robust performance of Random Forest and Support Vector Machine across different feature sets, underscoring their efficacy in authentication tasks.