Employing smartwatch for enhanced password authentication

Abstract

This paper presents an enhanced password authentication scheme by systematically exploiting the motion sensors in a smartwatch. We extract unique features from the sensor data when a smartwatch bearer types his/her password (or PIN), and train certain machine learning classifiers using these features. We then implement smartwatch-aided password authentication using the classifiers. Our scheme is user-friendly since it does not require users to perform any additional actions when typing passwords or PINs other than wearing smartwatches. We conduct a user study involving 51 participants on the developed prototype so as to evaluate its feasibility and performance. Experimental results show that the best classifier for our system is the Bagged Decision Trees, for which the accuracy is 4.58% FRR and 0.12% FAR on the QWERTY keyboard, and 6.13% FRR and 0.16% FAR on the numeric keypad.