Abstract: There are general problems in the study of student performance prediction, such as simple data structure, shallow linear layer of learner, etc. Therefore, considering the temporality of students’ historical performance and the forgotten features during learning process, a LSTM network was introduced to model the state of students’ knowledge structure. The features of emotion and behavior were integrated to predict the academic performance through fully connected neural networks. Experimental results show that the method can significantly improve the accuracy of prediction of academic performance. At the same time, a method for judging the main influencing factors of performance is further proposed.

Key words: LSTM, academic performance prediction, student emotion, student behavior