Predicting personality traits from touchscreen based interactions

Human influence factors can have a strong impact on how users perceive the quality of a given system. Traditional measures such as questionnaires can be a time consuming and partially invasive means to assess human influence factors like personality. In this paper, we investigate whether an individual's personality traits can be classified based solely upon the characteristics of their touchscreen usage. We record the tablet input of 75 subjects using a cognitive training application. The application requires the user to spell words by tapping on letters and dragging them to their denoted position on the screen. The task is presented in two conditions: in one with normal functionality of the program, the other with the usability of the application impaired. The subject's personality is measured with the NEO-FFI questionnaire and captured in the five dimensions of the five-factor model (Big 5). The personality scores and 68 features (touch-behavior and task-performance), as well as statistical metrics derived from them, are fed to a number of classification algorithms. Our results show that in a binary classification task with normal usability, high levels of ‘Neuroticism’ can be distinguished from low levels with a mean accuracy of 66 percent. The extent to which a personality dimension can be predicted is dependent on the usability of the test system. In a normal setting, ‘Extraversion’ can be predicted with an accuracy of 61 percent. However, with impaired usability, the prediction rises to an average of 67 percent. The experiment shows that records of touchscreen interactions allow for the prediction of personalities significantly better than random. The study of the human influence factor personality and its relation to perceived quality would be facilitated by using touchscreen interaction data as a fast and easily accessible estimate of a user's personality.