Determining Patterns in Computer Based Long Text Development Using Keystroke Dynamics

Abstract

This research discusses the ability to authenticate individuals using keystroke data in computer based text development. Most of previous keystroke dynamic authentication researches have only considered keystroke data of some pre-generated text samples, to find keystroke patterns which are unique to individuals. However, computer based text development is quite different than type writing a given pre- generated text, due to the effort it requires to generate texts related to different genres. This research is conducted to find whether there exist unique keystroke patterns related to individuals across different genres. The keystroke data related to four main genres (Descriptive, Narrative, Argumentative, Expository) and two other special tasks (copy-write text and memorised text) are considered in this study. Keystroke data generated by 40 individuals have been considered with regard to these tasks. Both Micro and Macro level features are used separately for classification with Random Forest Classifier and Support vector machine classifier. Results prove that, there exist unique keystroke patterns for individuals, regardless of the genre of the text they have developed. As another part of this research, the ability to classify the text into its related genre, using keystroke data, is evaluated. Results suggest that keystroke data can be classified into different genres accurately regardless of the individual who has developed it. The results of this research can be applied for unique individual authentication and intrinsic plagiarism detection in computer based long text development.