A trio of Indian boffins have studied the use of smartphone accelerometers as biometric sensors and concluded they could be a handy way to identify users.
Unlike the collaboration between American and Hong Kong researchers who want “who are you?” for ad-tracking, the National Institute of Technology, Karnataka boffins' research recommended using the accelerometer-ID to secure your mobe.
The researchers collected accelerometer data from ten users carrying a Galaxy J1 in their pockets, with a focus on how the test subjects walk, and sampled 50 times a second.
That yielded data whose physical inputs – how frequently your feet hit the ground, for example, and how hard – depend on who's doing the walking.
From that, the researchers went conducted the usual processes of feature extraction, modelling, and analysis to work out how well the phone's understanding of gait can be turned into a useful identifier.
A final sample of 31 features, with 3,600 examples in the dataset, suggested this method could be an effective way to identify users: across the ten individuals in the test, the researchers' model accuracy ranged from a worst performance of 93.85 per cent up to “subject 6”, which the model got right 99.7 per cent of the time.
The best-performed modelling algorithm of the four the researchers tested was the Random Forest Classifier, which averaged 96.79 per cent accurate identifications.
Their study, published at arXiv, is under consideration for next year's IEEE National Conference on Communications. ®