Research On The Prediction Model Of Students' Academic Level Under Blended Teaching

Blended teaching combines the advantages of online teaching and offline teaching,breaking through the limitations of time and space to a certain extent,which reflects the transformation of the "student-centered" teaching paradigm.With the expansion of the blended teaching,how to analyze and dig out useful information through the characteristics of learners' learning behavior to help learners improve their learning efficiency and improve teachers' teaching standards and quality has become an urgent problem to be solved.Focusing on the above problems,this paper constructs an academic prediction model that integrates learning behavior analysis-feature selection-stacking which is based on the analysis and mining of the learning behavior data generated by the students in the process of blended teaching.The main work is as follows:First and foremost,most scholars are committed to curriculum design and improvement,ignoring the analysis of student behavior and the prediction of academic performance in blended teaching research.Therefore,this paper analyzes the underlying behavior patterns which is based on the datas of learning behavior and the factors related to learning behavior and academic performance,according to the behavior data generated by students in the process of blended teaching,including classroom and after-class behavior data.Secondly,this paper innovatively proposes the application of Stacking fusion prediction model in blended teaching.In the process of constructing the Stacking fusion model,decision trees,random forests,and extreme gradient boosting are used as models in order to cast off the influence of the accuracy of the base model on the fusion model.Model evaluation involves selecting two base models which are suitable for fusion.According to experimental analysis.the prediction accuracy rate is relatively low obtained by directly using the data,so the model with the better performance of the two models is used to rank the feature importance of the data,and combining feature correlation in order to conduct feature selection and filter redundant features.After feature selection,experiments are performed on random forest,extreme gradient boosting and Stacking fusion model in turn,and the model parameters are tuned during this process.The experiment shows that the accuracy of the model,compared with the original random forest,is increased by 3.6%,the limit gradient is increased by 2% and the Stacking model is also increased by 0.8% with the combination of feature selection and Stacking fusion model.Therefore,choosing characteristic and base model with better performance for feature behavior analysis and selection can ensure the accuracy of the algorithm.Finally,constructing personal portraits of the students include the following steps: setting up tag collections which are based on various types of behavior data and basic information of the students,analyzing various types of data of the students,monitoring abnormalities in students' academic performance and providing timely warnings so as to provide personalized services for teachers and students.