Designing, Implementing, and Evaluating a Software AI Module for Security in Smartphones

Abstract

Cybersecurity has become increasingly important as technology continues to advance and play a larger role in our daily lives. Accessing sensitive information like one’s medical records, bank account login, and home address could be done with the tap of a few applications once a user’s smartphone password or personal identification number (PIN) is breached. A previous study protects the privacy of a user’s sensory data by using deep learning algorithms to learn a user’s normal pattern of behavior. The study also uses prediction error distribution to enhance detection accuracy in a hardware module known as Smartphone Impostor Detector. In this paper, we train and test LSTM networks with the same architecture and dataset discussed in [3] on both a computer program and smartphone application. We also emulate the Smartphone Impostor Detector (SID) in software to test the physical parameters of the hardware module and compare its performance to the computer and smartphone software. Experiments show that performances across the computer, smartphone, and SID emulation are almost identical in terms of classification metrics. However, the smartphone application turned out to be significantly faster than both the computer software and SID emulation software.