Feature Importance Evaluation of Smartphone Touch Gestures for Biometric Authentication

In this work, we present a method of feature selection for smartphone touch gesture classification. Touch gestures, also known as touchscreen features are used as behavioral attributes with machine learning classifiers to implement authentication systems for smartphones. We propose to use a publically available dataset and perform a feature scoring with the extreme gradient boosting (XGBoost) algorithm to select the most relevant features. We carried out two experiments: in the first one, we used a vector of 30 features for the classification and we performed feature ranking. In the second experiment, we used a subset of 7 features based on the ranking given by the XGBoost algorithm. Classification results are evaluated with the state of the art approaches. We achieved an accuracy of 99.41% using only a feature vector of 7 variables, this demonstrates that touchscreen features contain relevant information about the human identity and could be used for biometric authentication.