Gender Detection with Smartphone Motion Sensors Using Convolutional Neural Networks
Erhan Davarcı, Emin Anarım · 2021
In the literature, there are many studies showing that user characteristics such as age and gender can be obtained by using sensor data from smart devices. Behavioral differences in users' interactions with smart devices make these studies effective. In this study, it is shown that the gender of the user can be recognized using data from the accelerometer sensor on smartphones. In this context, we develop an Android application and collect data from 120 users interacting with their devices while walking. Then, we extract features in order to analyze the effects of differences in users' touching and gait behaviors on sensor readings. For gender detection, machine learning methods are firstly investigated and several classification algorithms are applied. In addition, we develop a Convolutional Neural Network (CNN) model and test it on our dataset. Consequently, we obtain an %88.3 success rate with CNN and also show that the CNN model outperforms other machine learning methods.