Eye-Gaze-Based Intention Recognition for Selection Task by Using SVM-RF

This paper focuses on the problem of intention recognition in eye-gaze-based interaction. The user’s intention could be divided into two types: selection and non-selection. In this study, a within-group experimental design was designed to complete the target letter selection task through eye-gaze-based interaction. Python is used to develop an experimental software for better flexibility in recording data. The SVM-RF model has been built and compared with other algorithms such as SVM, RF, etc. The importance weight of different eye movement behavior features on the accuracy of intention recognition has been analyzed by comparing the accuracy of the model through the permutation feature importance method. The results indicated that the SVM-RF model had a prediction accuracy of 94.3%, which can effectively predict the user’s selection intention and provides reference value for the future development of eye-gaze-based interaction technology.