Research On Grade Prediction Method Based On Student’s Online Learning Behavior

Because online learning breaks the restrictions of time and space,reasonably integrates learning resources and provides great convenience for students’ learning.The large number of people participating in online learning and flexible learning methods make it difficult for teachers to master the learning situation of each student and accurately identify the students at risk in learning.Student grade prediction can realize the evaluation of learning effect through the analysis of students’ relevant data,so as to achieve the purpose of identifying risk students.At present,student grade prediction algorithms are mainly studied from traditional machine learning and deep learning.Among traditional machine learning algorithms,ensemble learning algorithms have achieved good prediction results in the field of grade prediction.However,the existing ensemble learning algorithms in grade prediction are not ideal in efficiency and scalability when the dataset has a large amount of data and a high feature dimension.Among deep learning algorithms,recurrent neural networks can often achieve better prediction results in the field of grade prediction.However,due to the flexibility of learning methods in online education,students’ learning is often not constrained by time,so,the sequential information extracted from the sequence of students’ learning behaviors cannot fully and correctly reflect the students’ learning status.In response to the above problems,this paper studies them respectively,and the specific contents are as follows:(1)Aiming at the problems existing in grade prediction based on traditional machine learning,this paper introduces the LightGBM algorithm(Light Gradient Boosting Machine)to build a student grade prediction model.The grade prediction model can efficiently and quickly process online education datasets with large amounts of data and high feature dimensions.A large number of experiments have been carried out on the public online education data set.The experimental results show that,compared with other grade prediction models based on traditional machine learning algorithms,the grade prediction model can not only achieve more accurate student grade prediction,but also has a significant increase in training speed.(2)Aiming at the problems existing in grade prediction based on deep learning,this paper proposes a new network structure GDPN(GRU&DNN Prediction Network)to build a student grade prediction model.The grade prediction model firstly uses a recurrent neural network to extract the sequential information of learning behavior from the behavior sequence learned by students,and secondly uses a deep neural network to extract the overall information of learning behavior from the students’ overall learning behavior data,and finally uses the connect layer to combine the two kinds of information,and a simplified feedforward neural network attention mechanism is used to quickly and efficiently assign appropriate weights to the extracted information.The grade model can extract both sequential information and overall information in students’ learning behavior,so as to achieve more accurate grade prediction.A large number of experiments are carried out on the public online education data set,and the experimental results show that the prediction model has achieved better results in various evaluation indicators compared with other deep learning-based grade prediction models.