DAKOTA: Continuous Authentication with Behavioral Biometrics in a Mobile Banking Application

Behavioral biometrics focus on identifying patterns in human activities or behaviors. Tracking and identifying interaction behavior of a user with a computing device is relatively a novel approach for user authentication on computer systems. In this paper, we present a continuous authentication scheme (DAKOTA) proposed for a mobile banking application. The currently used banking application is modified to track the behavioral biometrics of the client. A data logger is added to collect touch screen data while a user interacts with the application, as well as sensor data from accelerometer, gyroscope and magnetometer to capture the hand movements of the user. 30 volunteers participated for initial data collection. A one-class SVM model is trained for authenticating each user. Classification results show that, we can achieve 79% correct user identification, 13% false acceptance rate and 11.5% equal error rate. We also implemented an end-to-end authentication scheme, where data are collected on the client while using the mobile banking application and transferred to a server where the features are extracted and the SVM model for the particular user is invoked. We also present the resource usage results of DAKOTA in comparison to the original banking application, in terms of energy, CPU and memory as well as delay and data size.