Identifying Users Based on Use of Input Devices

Abstract

The need for other security measures besides secret keywords has highlighted the possibility to base a biometric security system on input device patterns. This report explores the feasability of using mouse and keyboard biometry in an online security system. After examining prior work in this area, an online test environment was set up where subjects could remotely complete experiment sessions by interacting with a web-site. Artificial Neural Networks (ANN), k-Nearest Neighbor (k-NN) and Self Organizing Maps (SOM) were combined into profiles which predicted sample validity. A final system accuracy of 0.92% False Accept Rate (FAR) and 43.16% False Reject Rate (FRR) was achieved. The best profile in the system achieved a FAR of 0.8% and a FRR of 0.0%. When one of the 19 subjects was browsing the online test environment, the system could with an accuracy of 54.1% correctly identify the user. Examining the results, it is clear that data of a higher quality is needed and alternative collection methods should be explored. It seems feasible to recognize a registered user solely based on keyboard, mouse and browser variables.