Exploring Internal Correlations in Timing Features of Keystroke Dynamics at Word Boundaries and Their Usage for Authentication and Identification

The internal correlation of timing features inside word boundaries of free text sessions are studied, together with the classification performance in both authentication and identification tasks, using three publicly available datasets. Increasing distance between keystrokes was not found to be predictive of decreased correlation, even when individual users are considered and not the dataset as a whole. Considering five or more letter words, classification performance ranged 75%–90%+ for authentication and 60%–90%+ for identification tasks, with Random Forest classifier performing the best, and the simple k-nearest neighbours (k-NN) the worst. All the conclusions and observations generalised to the three datasets and its subsets, and particularly to both Spanish and English languages.