HANDS: enHancing Academic performaNce via Deep foreSt

Student performance prediction plays a critical role in numerous educational scenarios, e.g., academic early warning and personalized teaching. A large body of researches have been developed for better learning gains through leveraging plentiful student-related information, such as students' historical course grades and demographical data. Recent years, campus smartcards have been widely used in most of Chinese universities, and a large amount of students' behavioral data can be recorded in an unobtrusive style, which provides a new perspective for us to predict students' academic performance. Different from most of traditional approaches, in this study, we try to exploit students' campus behavioral data, and present a novel approach for student performance prediction, namely as HANDS (enHancing Academic performaNce via Deep foreSt), which introduces decision tree-based deep learning method into establishing the predictive models. Particularly, benefiting from end-to-end learning style of deep learning, the proposed HANDS method can automatically extract students' behavioral characteristics and predict their academic performance. And thus it can save expensive human labor cost, compared with handcrafted feature-based approaches. Experimental results on a real-world data set demonstrate the superiority of our approach over the state-of-the-art methods.