Feature Selection for Android Keystroke Dynamics

Abstract

Keystroke Dynamic Authentication is the way of authenticating users by analyzing their typing rhythm and behavior. While key hold time, inter-key interval time and flight time can be captured on all devices; applying Keystroke Dynamic Authentication to mobile devices allows capturing and analyzing additional keystroke features like finger area on screen, and pressure applied on the key. This paper aims to reduce the number of captured features without affecting the efficiency of the user prediction. For this purpose, we used a benchmark dataset and implemented 3 different filter feature selection methods to sort the features by their relevance. Sets of different sizes were created and tested against classification methods.