Active Authentication Using Touch Dynamics

Mobile devices are holding an increasing amount of private and financial information that security has not been able to keep up with. It has been found that 15% of PIN numbers have the same combination of numbers, and all are susceptible to “shoulder surfing”. Others have tried to remedy this issue with fingerprint scanners, facial recognition, or other biometric features, but these still leave the device vulnerable after authentication. Active authentication can continuously authenticate a user without the need for a password from the user. We achieve this by analyzing a user's behavioral biometrics while using their device. We use the Hand Movement, Orientation, and Grasp (HMOG) dataset which includes sensor and input data from 100 users over 24 sessions. Unlike other studies, we use the raw sensor data without any augmentation or feature extraction. We test this with the application of four different classifiers: Long Short-Term Memory Recurrent Neural Network (LSTM-RNN), Random Tree, Bayesian Network, and J48 Decision Tree.