Performance Analysis for a Touch Dynamic Authentication System with Reduced Feature Set Using Neural Networks

Increase in the usage of smart phones increases the burden for the users to memorize many passwords. It has also increased the need for stronger or enhanced authentication mechanisms. Littlemore innovation labs are providing paperless solution for conducting examinations using touch pad that is believed to be a promising technology in future. This paper explores the suitability of using touch dynamics as an additional level of security during authentication. Both security and performance concerns are investigated. A security application developed can be used on a touch screen device capable of imitating someone's typing characteristics. When the user starts interacting with the device, the developed application starts capturing the behavioural features given by the user's swipe. Thirty different features are captured, from which, most prominent features are identified by depending on the usefulness of the feature. Good results with the reduced feature set are obtained, thereby improving the performance of the system using neural network techniques. The results show that touch dynamics on a smart phone are more durable against certain attacks on a personal computer.