Classification of Input Devices Using Cursor Trajectories

With the immense popularity of the internet and the much lower cost for online surveys compared to in-person interviews, many surveys are now conducted online. One approach to identify whether respondents are having difficulty answering a question is by analysing cursor trajectory data. There are many different input devices, for example computer mouses and touchpads. Respondents’ behaviour and thus their cursor trajectories might vary depending on which device is used. As a first step toward tackling this issue, this thesis tries to differentiate between trajectories produced by using a mouse and those made by a touchpad. There are two classification methods used here; one is using a simple logistic regression model to try and classify the input device, the other is a tree-based approach, more specifically, a Random Forest model. Furthermore, two models with different features are being fitted to the data: a baseline model using features based on previous research as well as a full model utilising several other, new features. Both methods and models offer a very limited capability to classify respondents’ input devices. A comparison between them shows that the full Random Forest model exhibits the best performance of the methods used. Finally, measures to improve the classification performance and the question whether a differentiation between the input devices is necessary at all, are discussed.