Gender recognition in smartphones using touchscreen gestures

Abstract

This paper presents an approach for gender recognition in smartphones using touchscreen gestures performed by the user. The primary behavioral data comprising readings from the accelerometer, gyroscope, and orientation sensors are acquired while the user interacts with the touchscreen device. These measurements are further enriched by deriving a secondary set of gesture attributes such as swipe length and point curvature. The GIST descriptor-based features are then extracted from two-dimensional maps of the gesture attributes. Finally, a k-nearest neighbor (k-NN) classifier recognizes the user's gender based on a subset of features identified through feature selection. We have evaluated the performance of the proposed approach on two datasets, which consist of 2268 touch gestures from 126 subjects, collected using two different touchscreen devices. Our experiments show that the approach achieves higher gender classification accuracy compared to the existing method. In addition, the performance of our approach is consistent as it provides classification accuracy of 93.65% and 92.96% on the first and second datasets, respectively when multiple gestures are combined for gender recognition. Our study demonstrates that holistic image features considered in this work provide reliable information for smartphone-based gender classification.