Unlocking Personal Characteristics: Harnessing Keystroke Dynamics for Identification on Mobile Devices

Campbell, James (2024) Unlocking Personal Characteristics: Harnessing Keystroke Dynamics for Identification on Mobile Devices. Doctoral thesis, University of East Anglia.

Preview

PDF Download (2MB) | Preview

Abstract

The widespread adoption of mobile devices has led to rapid technological advancements. These advancements have made the addition of motion sensors, such as accelerometers and gyroscopes, critical components of a mobile device.

These sensors offer novel methods of identification, which can be employed to harness this technology providing a transparent, yet more secure process. One such method, keystroke dynamics, has been present across more traditional physical devices, such as keyboards for a long time, and has more recently bridged across into virtual keyboards and mobile devices.

This research explores the potential of keystroke dynamics, augmented by motion sensor data, to infer personal characteristics such as name, age, gender and handedness with high accuracy. Three key research questions are addressed to form the basis of the thesis:
1. To what extent can keystroke dynamics be utilised in order to infer a person's name on a mobile device?
2. What effect does the inclusion of accelerometer and gyroscopic data alongside keystroke dynamics have on the ability to successfully infer a person's name and soft biometric features on a mobile device?
3. To what extent does the volume of data per user help to improve the accuracy of the prediction of name and soft biometric features?

An innovative approach to identification is presented alongside key contributions of the work, which include:
_ Novel approach to identity data using smartphone motion sensors.
_ A new methodology and experimental approach (including data capture framework).
_ Bespoke data sets of motion data that can be anonymised and shared with the wider community.
_ A novel algorithm for predicting a letter utilising smartphone motion sensors.

A comprehensive methodology is employed, combining data collection across multiple studies with detailed machine learning and manual analysis to provide high accuracy scores. The results demonstrate that name can be inferred with significant accuracy (83.20%), whilst also achieving high accuracy rates for age (87.74%), gender (82.56%) and handedness (93.18%). Furthermore, increasing the volume of collected data per user led to further improvements in prediction accuracy.

The findings highlight the potential for sensor-enhanced identification, offering improved security while maintaining usability and transparency. This research contributes to the field of behavioural biometrics by advancing the understanding of sensor-enhanced mobile-based identity inference, paving the way for more secure, transparent and seamless methods of identification.

Item Type:	Thesis (Doctoral)
Faculty \ School:	Faculty of Science > School of Computing Sciences
Depositing User:	Chris White
Date Deposited:	02 Apr 2025 08:51
Last Modified:	02 Apr 2025 08:51
URI:	https://ueaeprints.uea.ac.uk/id/eprint/98919
DOI:

Downloads

Actions (login required)

View Item