A Behavioral Biometrics User Authentication Study Using Motion Data from Android Smartphones

Abstract

This is a study of the behavioral biometric of smartphone motion to determine the potential accuracy of authenticating users on smartphone devices. The study used the application Sensor Kinetics Pro and the Weka machine-learning library to analyze accelerometer and gyroscope data. The study conducted three experiments for the research. They were conducted in spring 2015, fall 2015, and spring 2016. The final experiment in spring 2016 used six Android-based smartphones to capture data from 60 participants and each participant performed 20 trials of two motions: bringing the phone up to eye level for review, and then bringing the phone to the ear, resulting in 1200 runs. The resulting sensor datasets were used for machine learning training and testing. The study used filtering data to remove noise, and then aggregated the data and used them as inputs to the Weka Machine Learning tool. The study used several machine classification algorithms: the Multilayer Perception (MLP), k-Nearest Neighbor (k-NN), Naïve Bayes (N-B), and Support Vector Machine (SVM) machine learning classification algorithms. The study reached authentication accuracies of up to 93% thus supporting the use of behavioral motion biometrics for user authentication. Preliminary studies with smaller numbers of participants in spring 2015 and in fall 2015 also produced 90%+ authentication accuracy.