Democritus University of Thrace (DUTH)
Κατηγοριοποίηση χρηστών δια μέσου δυναμικής της πληκτρολόγησης, για ταυτοποίηση υπόπτου
Abstract
dc:descriptionThe study subject of this PhD thesis is the development of a method that is able to reveal part of identity of an unknown user who communicates through Internet. A user can hide behind the anonymity and this offer an ensuring that a leakage of personal and critical data will not happen. However, when this anonymity used to conceal illegal activity, then it becomes necessary the protection of unsuspicious users from malicious attacks which endangering security associated with personal, social and financial issues.Although they have been proposed various ways to succeed something like this, each of them has at least one weakness and therefore it cannot be used as general solution. In contrary, in this research, the necessary data comes from times of events that take place on the keyboard. The most important advantages of the proposed method are three. First, it is studied the user behavior during the text typing, which is the most prevalent media of the users communication on Internet. Second, the whole research was conducted based on QWERTY keyboard, which is the most common keyboard layout and it can be found both on "traditional" devices, like desktops and laptops, and "modern" devices, like tablets and smartphones, where has the virtual form. Finally, third, the parameters that used to reach the goals doesn't related to specific language, because they have to do with how a user types and not with what he/she types. The consequence of all these is that the proposed method can be generally applied, since it concerns the main communication media of users, the most frequently occurring text composition device and it is language independent.The computer science field that deals with the study of detailed timing information of keyboard events is called keystroke dynamics and it is almost exclusively used to user authentication. The keystroke dynamics parameters that can be exploited for extracting an outcome are too many, but each of them includes small amount of information. In this research, keystroke dynamics was used to classify users according to their characteristics and the parameters that were exploited were the keystroke durations and the digram latencies.Having as the purpose the correct prediction of an unknown user characteristics, so that to reveal part of his/her identity, systems which have as input data that came from the way a user types and extract a result that classify the user into a category were created. These systems were based on classifiers, which were chosen by evaluating the correct prediction percentage, the operating speed and the performance stability. More specifically, they were used Bayesian classifiers, Euclidean distance classifiers, Manhattan distance classifiers, support vector machines (SVM), multilayer perceptrons (MLP) and radial basis function networks (RBFN).The necessary data for the study were obtained in two different ways. First, by asking volunteers’ groups to type specific text while they were recorded, this is the fixed text approach. Second, by asking volunteers to use their computers during everyday activities, this is the free text approach. With fixed text data was followed a procedure, characterized as the first stage of research, in which was studied the user classification according the typing device and according their gender. With free text data was followed another procedure, characterized as the second stage of research, in which was studied the user classification according their dominant hand, according their age group, according their educational level and according computer daily usage.The results show that the user classification through keystroke dynamics is feasible. The correct prediction percentages considered satisfying and useful conclusions are drawn to continue research in this direction.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2017
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tsimperidis, Ioannis
- Τσιμπερίδης, Ιωάννης
Subjects
dc:subject × 10Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/40524