The habitual manner of interacting with a smartphone while making text entry, also known as keystroke dynamics, can be used for age-group and gender prediction. Knowing what type and amount of the keystroke dynamic features used will improve the effectiveness of the predictive biometric system. While such efforts have been made in the area of predictive keystroke dynamics using computer keyboards, literature on the same topic using touchscreen smartphone virtual keyboards have been limited. In this paper, keystroke dynamics data of 50 individuals have been acquired using an open-source data software application on an Android smartphone. A total number of 21 commonly used keystroke dynamics features were extracted from the raw data. The collected data was used in training a Random Forest algorithm using four different training sample sizes while the remaining portion of the data was used for classification. The algorithm was then used to determine the importance of 21 different keystroke dynamics features. The results showed that each features offers varying degree of importance in age-group and gender predictions. Understanding the usefulness of each feature will enable the selection of the most efficient amount of input features required for age-group and gender prediction tasks.
Keywords: Biometrics, Machine Learning, Keystroke Dynamics, Mobile Phones