HCI practical course, University of Stuttgart
On the basis of the EMVA dataset, we discuss the feasibility on the topic of Anticipating Application Switching during mobile device interactions. Previous work on the related anticipating application switching is based on the features from mobile device usage logs, sensor data and contextual data, and finished the prediction task mostly with the help of wearable equipment which is in daily mobile phone usage scenarios not necessary. In our study, we focus on the features of user behavior like gaze data and screen touch frequency during the daily interaction with the mobile device. Feature extraction works on with the mobile phone usage features from interactive data, and features of attentive behaviour from the front-facing camera. By OpenFace we process the video and generated the related gaze data. After that we implement a suitable machine learning model for the classification task, evaluate the performance and analyse the generated results. We demonstrate that method can anticipate the application switching, i.e. which app the user intends to switch, and may help to develop tools or plug-in that increase productivity while minimising possible distractions from the environment. We also discuss about the remaining challenges for improving the performance in the future work.