Hybrid Architecture for Gender Recognition Using Smartphone Motion Sensors
Erhan Davarcı, Emin Anarım · 2021 29th European Signal Processing Conference (EUSIPCO) · 2021
Motion sensor data in smart devices can be used as a side-channel to capture user's behavioral biometrics. In this paper, we investigate the feasibility of using smartphone motion sensors to detect the gender of the user. The main idea behind our study is based on behavioral differences of male and female users in touching and holding the smartphones. In order to implement our method, we collect data from 100 subjects while they are performing different activities like sitting, standing and walking. Our experiments point out that tapping behaviors are very discriminative for gender recognition but their significance decreases while users are walking. To address this issue, we also implement user activity detection and propose a hybrid model to predict gender in different user activities. In this context, user activity is firstly detected and then gender is predicted correspondingly. Consequently, we show that gender recognition is implicitly performed with a success rate of 85% in sitting and standing activities, whereas 83% success rate is acquired in walking scenario.