The field of learning analytics has been limited by its frequent dependence on learning logs created by students while learning. Most of the research has dealt with the relationships between learning during a course and the achieved results. Although students' in-class behavior affects learning achievement, this remains a challenging aspect to study because of the difficulty in collecting relevant data. There is little research wherein in-class behavior is directly analyzed. We have developed a tabletop device that is able to detect the weights and contact areas of objects placed upon it. In this paper, we report a viable classification scheme for student in-class behavior, and demonstrate the potential for machine learning to automatically classify behaviors. The results showed the approach taken by Support Vector Machine (SVM) achieved 66.69% classification accuracy and the student's in-class behavior detection through the tabletop device might be feasible.