Key points are not available for this paper at this time.
Mouse dynamics, information on user's interaction with a computer mouse, are in vogue in machine learning for purposes such as recommendations, personalization, prediction of user characteristics and behavioral biometrics. We point out a blind spot in current works involving mouse dynamics that originates in underestimating the gravity of the characteristics of the mouse device and configuration on the data that mouse dynamics are inferred from. In a controlled study with N=32 participants, across three kinds of mouse interaction activities, we collect data for mouse dynamics utilizing a variety of mouse parameter configurations. We show that mouse dynamics commonly used in studies can be significantly altered by differences in mouse parameters. Out of 108 evaluated mouse dynamics metrics, 95 and 84 are affected between two conducted studies. A machine learning model's performance can be warped by the mouse parameters being used. We demonstrate on a prediction task that mouse parameters cannot be approached uniformly and without consideration. We discuss methodological implications - how mouse dynamics studies should account for the diversity of mouse-related conditions.
Kuric et al. (Mon,) studied this question.